Detecting protein complexes in living cells from laser scanning confocal image sequences by the cross correlation raster image spectroscopy method.
ABSTRACT We describe a general method for detecting molecular complexes based on the analysis of single molecule fluorescence fluctuations from laser scanning confocal images. The method detects and quantifies complexes of two different fluorescent proteins noninvasively in living cells. Because in a raster scanned image successive pixels are measured at different times, the spatial correlation of the image contains information about dynamic processes occurring over a large time range, from the microseconds to seconds. The correlation of intensity fluctuations measured simultaneously in two channels detects protein complexes that carry two molecules of different colors. This information is obtained from the entire image. A map of the spatial distribution of protein complexes in the cell and their diffusion and/or binding properties can be constructed. Using this cross correlation raster image spectroscopy method, specific locations in the cell can be visualized where dynamics of binding and unbinding of fluorescent protein complexes occur. This fluctuation imaging method can be applied to commercial laser scanning microscopes thereby making it accessible to a large community of scientists.
- SourceAvailable from: PubMed Central[Show abstract] [Hide abstract]
ABSTRACT: Transient receptor potential vanilloid 4 (TRPV4) channels are Ca(2+)-permeable, nonselective cation channels expressed in multiple tissues, including smooth muscle. Although TRPV4 channels play a key role in regulating vascular tone, the mechanisms controlling Ca(2+) influx through these channels in arterial myocytes are poorly understood. Here, we tested the hypothesis that in arterial myocytes the anchoring protein AKAP150 and protein kinase C (PKC) play a critical role in the regulation of TRPV4 channels during angiotensin II (AngII) signaling. Super-resolution imaging revealed that TRPV4 channels are gathered into puncta of variable sizes along the sarcolemma of arterial myocytes. Recordings of Ca(2+) entry via single TRPV4 channels ("TRPV4 sparklets") suggested that basal TRPV4 sparklet activity was low. However, Ca(2+) entry during elementary TRPV4 sparklets was ∼100-fold greater than that during L-type CaV1.2 channel sparklets. Application of the TRPV4 channel agonist GSK1016790A or the vasoconstrictor AngII increased the activity of TRPV4 sparklets in specific regions of the cells. PKC and AKAP150 were required for AngII-induced increases in TRPV4 sparklet activity. AKAP150 and TRPV4 channel interactions were dynamic; activation of AngII signaling increased the proximity of AKAP150 and TRPV4 puncta in arterial myocytes. Furthermore, local stimulation of diacylglycerol and PKC signaling by laser activation of a light-sensitive Gq-coupled receptor (opto-α1AR) resulted in TRPV4-mediated Ca(2+) influx. We propose that AKAP150, PKC, and TRPV4 channels form dynamic subcellular signaling domains that control Ca(2+) influx into arterial myocytes.The Journal of General Physiology 05/2014; 143(5):559-75. · 4.57 Impact Factor
- [Show abstract] [Hide abstract]
ABSTRACT: The structure of cell membranes has been intensively investigated and many models and concepts have been proposed for the lateral organization of the plasma membrane. While proteomics and lipidomics have identified many if not all membrane components, how lipids and proteins interactions are coordinated in a specific cell function remains poorly understood. It is generally accepted that the organization of the plasma membrane is likely to play a critical role in the regulation of cell function such as receptor signalling by governing molecular interactions and dynamics. In this review we present different plasma membrane models and discuss microscopy approaches used for investigating protein behaviour, distribution and lipid organization.Molecular Membrane Biology 07/2014; · 1.73 Impact Factor
- [Show abstract] [Hide abstract]
ABSTRACT: Fluorescence Correlation Spectroscopy (FCS) is a widely used technique in biophysics and has helped address many questions in the life sciences. It provides important advantages compared to other fluorescence and biophysical methods. Its single molecule sensitivity allows measuring proteins within biological samples at physiological concentrations without the need of overexpression. It provides quantitative data on concentrations, diffusion coefficients, molecular transport and interactions even in live organisms. And its reliance on simple fluorescence intensity and its fluctuations makes it widely applicable. In this review we focus on applications of FCS in live samples, with an emphasis of work in the last 5 years, in the hope to provide an overview of the present capabilities of FCS to address biologically relevant questions.FEBS letters 04/2014; · 3.54 Impact Factor
Detecting Protein Complexes in Living Cells from Laser Scanning
Confocal Image Sequences by the Cross Correlation Raster Image
Michelle A. Digman,†Paul W. Wiseman,‡Alan R. Horwitz,§and Enrico Gratton†*
†Laboratory for Fluorescence Dynamics and Department of Biomedical Engineering, University of California, Irvine, California;‡Departments
of Chemistry and Physics, McGill University, Montreal, Quebec, Canada; and§Department of Cell Biology, School of Medicine, University
of Virginia, Charlottesville, Virginia
rescence fluctuations from laser scanning confocal images. The method detects and quantifies complexes of two different fluo-
rescent proteins noninvasively in living cells. Because in a raster scanned image successive pixels are measured at different
times, the spatial correlation of the image contains information about dynamic processes occurring over a large time range,
from the microseconds to seconds. The correlation of intensity fluctuations measured simultaneously in two channels detects
protein complexes that carry two molecules of different colors. This information is obtained from the entire image. A map of
the spatial distribution of protein complexes in the cell and their diffusion and/or binding properties can be constructed. Using
this cross correlation raster image spectroscopy method, specific locations in the cell can be visualized where dynamics of
binding and unbinding of fluorescent protein complexes occur. This fluctuation imaging method can be applied to commercial
laser scanning microscopes thereby making it accessible to a large community of scientists.
We describe a general method for detecting molecular complexes based on the analysis of single molecule fluo-
The past two decades have produced a revolution in optical
microscopy. Nonlinear excitation, stimulated emission, and
more recently, single molecule imaging, for example, have
pushed the limits of optical resolution to new frontiers (1–4).
Despite these advances, however, a need remains for a robust
method for detecting protein complexes in living cells.
Because cellular processes are often localized and transient,
the ideal method would have high spatial resolution, and the
data should be acquired within the timescale of the biological
process under investigation.
Generally, the existence of molecular complexes is in-
ferred biochemically using coimmunoprecipitation and
then confirmed by fluorescence colocalization or FRET
(5,6). Colocalization, even at the super resolution achievable
with the most recent fluorescence methods, does not show
that the molecules of interest actually reside in a structural
complex. Whereas molecules residing within ~5 nm show
FRET under ideal situations, two different molecules within
a structurally defined complex, which contains several
different molecular species, may not be close enough for
FRET. Furthermore, molecules that do not reside in the
same structural complex but are near each other can exhibit
Fluorescence correlation spectroscopy, which is based
on dynamic colocalization, is an alternative approach (7).
However, fluorescence cross correlation methods are based
traditionally on the measurement of temporal fluctuations
at a single point in the cell. In this mode, measuring cross
correlated fluctuations at a single point is difficult to interpret
in living cells because of possible correlations due to move-
ment of macroscopic objects. In addition, it requires that the
observer choose a point of interest before the measurement
begins. Thus, a method is desired that can separate the
obvious correlations due to the movements of macroscopic
objects and at the same time provide a map of the location
of specific molecular complexes.
Confocal fluorescence microscopy has revolutionized the
biomedical field and allowed monitoring of biological
processes in live cells in 3D and in real time. However, the
wealth of information contained in the confocal image has
not been fully exploited to date. We show that we can deter-
mine molecular interactions directly in live cells from
confocal images. In general, intensity fluctuations are caused
by diffusion or binding/unbinding interactions of the protein
complex. The coincidence of fluctuations occurring at two
detection channels shows that the two proteins are part of
the same complex.
In this study, we exploit the raster-scan image correlation
spectroscopy (RICS) method that can analyze the diffusion
and binding dynamics of molecules in an entire, single image
rather than at single points on an image (8,9). We extend the
RICS approach to extract the spatial and temporal informa-
tion provided by the cross correlation between two different
types of proteins measured using two detection channels.
The basis of the RICS method has been described previously
(8). Briefly, in a raster scanned image, the fluorescence inten-
sity of different pixels are measured in a temporal sequence.
If molecules move on the timescale of the scan speed, which
is microseconds along each scan line and milliseconds
between lines, the spatial correlation function for the image
Editor: Alberto Diaspro.
? 2009 by the Biophysical Society
Biophysical Journal Volume 96 January 2009 707–716707
is affected by the movement of molecules from a previously
scanned pixel to the new pixel being scanned. When done on
each image of an image series, we can extract the time infor-
mation associated with diffusion processes and binding-
unbinding equilibria in different parts of an image over
a time window that includes most biological processes.
The RICS approach can be extended to pairs of molecules,
using two-color cross correlation, to measure the diffusion
of protein complexes, estimate the fraction of interacting
molecules, and determine the temporal and spatial distribu-
tion of these complexes. This method, which we call cross
correlation RICS (ccRICS), is generally applicable and can
be done using commercial, one photon, scanning confocal
laser microscopes (10). ccRICS differs in concept from other
image based correlation measurements, e.g., image cross
correlation spectroscopy (ICCS) (11). The later, for example,
correlates intensity fluctuations of pixels or pixel regions
among images in a sequence rather than that of the pixels
along the raster scan of a single image. Because of this differ-
ence, ICCS and other imaged based methods are limited to
changes on the timescale of seconds, whereas ccRICS
widens the time regime to processes as fast as cytoplasmic
In this study, we apply ccRICS to a set of adhesions mole-
cules that are known to associate in solution by coimmuno-
precipitation. We show that we can distinguish between
cytoplasmic diffusion and binding and generate maps of
molecular interactions and the dynamics of these interactions
across the cell. We also show that these molecules do not
reside in complexes when they are diffusing in the cytoplasm
but do reside in complexes in the vicinity of disassembling
adhesions. These molecules in these complexes show no
FRET indicating that in the complex the proteins are at
a distance at which FRET is not taking place. From these
observations, a model emerges in which adhesions assemble
semble by releasing complexes that rapidly disassociate.
These studies show the rich content of the ccRICS anal-
ysis. Interaction maps can be drawn among rapidly diffusing
complexes (in the ms-ms timescale), complexes undergoing
binding and unbinding interactions (in the ms to s timescale)
and changes in the spatial distribution of complexes in
response to stimulation that occurs in minutes. The method
is noninvasive, and the same cell can be observed for long
periods. The map of protein interactions and the dynamics
of the interactions cannot be obtained using the co-immuno
precipitation methods, and FRET is of only limited use in
studying complexes that contain many molecules.
The ccRICS approach
In the ccRICS experiment, data are acquired in two channels
simultaneously. The two channels can contain data from two
spectral band passes or two polarization directions. When
two channels are acquired simultaneously in the confocal
microscope, they need to overlap well without significant
spectral bleed through as discussed in the Methods section.
In the RICS analysis, we calculate the 2D spatial correla-
tion function (8). The mathematical operation consists of
multiplying the matrix corresponding to the image by itself
at different shifts in the x and y directions. For ccRICS the
two images to be multiplied come from the two channels.
Note that the ccRICS is nonsymmetric with respect to the
order of the channels:
GccRICSðx;jÞ ¼hI1ðx;yÞI2ðx þ x;y þ jÞi
The variables x and j represent spatial increments in the
x and y directions, respectively. The 2D spatial correlation
is computed more efficiently using FFT methods rather
than by the formula above (13).
Whereas the RICS correlation function is sensitive only to
fluctuations of the signals in the individual channel, the
ccRICS is different from zero only when the intensity fluctu-
ations of the signals in the two detection channels are corre-
lated. Of course, there is always bleed through of one
channel into another (generally the green into the red). In
the absence of other correlations, bleed through will give
100% correlation between the two channels.
The amplitude of the correlation at shift (0,0) shows the
magnitude of the autocorrelation and of the cross correlation.
For two uncorrelated species, the amplitude of the cross
correlation at shift 0,0 is proportional to:
f11f12hN1i2þðf11f22þ f21f12ÞhN1ihN2i þ f21f22hN2i2
f11f12hN1i þ f21f22hN2i
where hN1i and hN2i are the average number of molecules of
species 1 and 2 in the volume of excitation. f11, f12, (f21and
f22) are the fractional fluorescence intensities of species 1
(species 2) as seen by channel 1 and by channel 2, respec-
tively. f12, f21, represent the bleed through of channel 1
into 2 and vice versa.
The range of the cross correlation signal achievable is
restricted to be equal, at most, to the amplitude of the auto-
correlation (Gcc(0,0) < ¼ G1 or 2(0,0)). The minimum value
is more difficult to predict because in principle the amplitude
could be negative if there is anti correlation, which could
occur if there are dynamic processes in the protein complex
itself. For ‘‘static’’ protein complexes, the minimum value of
the amplitude of the cross correlation is given by the amount
of bleed through and by the amount of uncorrelated signal.
The uncorrelated fluctuation, which arises either from back-
ground fluorescence or from molecules that are uncorrelated,
has the effect of strongly reducing the correlation due to
Biophysical Journal 96(2) 707–716
708Digman et al.
Cell culture and protein transfection
Mouse embryonic fibroblasts (MEF) were grown at 37?C in a 5% CO2
humidifiedincubator.Thecells weretrypsinized,subcultered andtransferred
from a 35-mm tissue culture flask to a 25 mm, 6-well Falcon tissue culture
(Becton-Dickinson, Bedford, MA). They were grown to 50–80% confluence
and transfected with 1 mg of DNA (0.5 mg of DNA/protein for cotransfec-
tions) and 5 mg of Lipofectamine 2000 obtained from Invitrogen (Carlsbad,
CA). Vinculin, focal adhesion kinase (FAK), and paxillin cDNA were
ligated to EGFP or mCherry at the C-terminal end. The FAK mutant,
I937E/I999E was ligated to EGFP at the C-terminal end. After 24 h of trans-
fection cells were trypsinized and plated using high glucose DMEM media
(Pierce-Hyclone, Logan, UT) supplemented with 10% FBS and PEN/
STREP on MatTek (Ashland, MA) imagingdishes coated with 3 mg of fibro-
nectin from Sigma-Aldrich (St. Louis, MO) 1 h before imaging.
We used an Olympus FV1000 microscope with a 60? 1.2NA water objec-
tive (Olympus, Tokyo, Japan). The scan speed was set at 12.5 ms/pixel. The
scan area was 256 ? 256 pixels and ~100 to 200 frames were collected for
each sample. The corresponding line time was 4.325 ms and the frame time
was 1.15 s. The electronic zoom of the microscope was set to 16.3, which
corresponds to a region of 12.5 mm2. For the EGFP excitation, we used
the 488 nm line of the argon ion laser and for the mCherry excitation we
used 559 nm excitation. The power of the 488 nm laser was set at 0.5%
according to the power slider in the FV1000 microscope. When the slider
is set to 100%, the power at the sample was 0.7 mW; we verified that the
slider operated linearly in the range used. For the red laser, when the slider
was set to 100%, the power was 0.1 mW at the sample. The power of the red
laser was then changed to match the average intensity in the two channels.
Generally, the power in the red channel was <1.5%. Data were collected in
the pseudo photon counting mode of the Olympus FV1000 microscope. The
filters for the green and red emission channels have a nominal bandwidth of
505–540 nm and 575–675 nm, respectively. The overlap of the volume of
observation and excitation at the two colors of our experiments was tested
by imaging single 100 nm fluorescent beads carrying two colors simulta-
neously (yellow-green fluorospheres; Invitrogen). We imaged single immo-
bilized beads using a z-stack with images acquired every 500 nm in the z
direction. We found that in the FV1000 microscope the center of mass of
the excitation volumes were coincident within 20 nm in the x and y direction
and within ~40 nm in the z direction in both channels.
We used the SimFCS program (Laboratory for Fluorescence Dynamics) for
RICS and ccRICS analyses. For the RICS analysis data were collected in the
256 ? 256 frame format. Fitting of the RICS functions was carried out
according to the equations for diffusion as described in Digman et al. (8)
and are presented below. The G(0,0) term, which contains the shot noise,
is omitted from the analysis. One of the features of the RICS analysis is
that slowly varying signals (fluctuations) can be removed from the calcula-
tion using a high-pass filter operation implemented by a moving average
operation (8). The moving average processing of the image stack removes
the spatial correlations due to the immobile fraction and the correlations
dueto slowly moving features in an image. The length of the moving average
determines the timescale of processes that are filtered out by this mathemat-
ical procedure. For the calculation of the scan analysis, a small region of
interest (64 ? 64 pixels) was systematically moved across the image by steps
of 32 pixels providing a partial superposition of the regions explored.
Subtraction of the immobile fraction
The RICS analysis consists of calculating the image spatial correlations.
This also contains the intensity correlations due to the image features.
Because we are only interested in the fluctuating part of the signal, the
average image is subtracted before calculating the spatial correlations. To
obtain the average image we collect several (~100–200) frames from the
image stack. The average of theseframes producesthe average image, which
is then subtracted, pixel-by-pixel, from each of the images of the stack. After
subtraction, the average difference has a value close to zero. To avoid
dividing by zero, when calculating the spatial correlation function, and to
properly normalize the RICS function after subtraction of the immobile frac-
tion, we add a number equal to the average of the average image to each
pixel of the subtracted images. If the image features vary slowly, due to
cell movement for example, we perform the subtraction operation using
only few frames of the stack chosen symmetrically around the image of
the stack that will be subtracted. The number of frames included in the local
average is determined by the moving average length parameter. If the
moving average length is small, i.e., only a few frames, slow variations
from frame to frame are effectively removed. By changing the length of
the moving average, we can specify the timescale of slow intensity fluctua-
tions that will be included in the fluctuation analysis. For example, using
a moving average of 10 (MAV10) all the fluctuations longer than 10 frames
(10 s in our case) areremoved.Using a movingaverageof 40 (MAV40)only
the very long fluctuations lasting 40 s or more are removed. Note that
this operation only affects the slow fluctuations that propagate from frame
to frame. The fast fluctuations that only propagate from pixel to pixel
or from line to line are not affected by the moving average subtraction
In the RICS approach, we calculate only the spatial correlations in one
frame. Bleaching or other intensity changes that propagate from frame to
frame do not affect, in principle, the RICS calculation. However, because
the frame-to-frame information is used to subtract the immobile (or quasi-
immobile) features of the image, the immobile subtraction algorithm will
not remove sudden changes in shape or position. These macroscopic
changes should not be confused with the correlation in position and intensity
due to point particles. Most of the methods based on fluctuations and using
camera acquisition are sensitive the changes of pixel intensity from frame to
frame. As such, these techniques are substantially different from the RICS
approach. In fact, the fast diffusive motion of small molecular aggregates
is averaged out due to the long exposure time of the camera and the pixel
to pixel correlation (in the same frame) cannot be used because each pixel
of the frame is acquired at the same time.
If one point in the image suddenly changes intensity due to bleaching,
binding, or blinking, these changes contribute to the RICS signal because
from pixel to pixel or from line to line the intensity fluctuation correlates
to neighboring pixels within the size of the point spread function. However,
if the immobile fraction removal algorithm is carried out with a very short
moving average length, the spatial effect of the local change in intensity
rapidly disappears from the image stack. For slow equilibria, the amplitude
of the correlation tends to decrease as the moving average length decreases.
Instead, if the binding-unbinding process is fast, the subtraction algorithm
will not influence the amplitude of the correlation. Fig. 1 shows simulations
of diffusion and binding–unbinding equilibria with different rates and the
effect on the RICS function. Fast diffusion always gives an elongated shape
along the x (fast) axis in the RICS function due to the probability to correlate
the same particle at a distance whereas slow binding equilibria give a round
shapebecausethe processoccursat specificlocations(Fig.1,left column).In
the central column of Fig. 1 we simulated particles binding in rapid (in the
ms timescale) equilibria to fixed locations. Because the dynamics occurs in
times comparable to the line time, when the position of the particle is visited
again at the next line, the intensity at this location appears to blink rapidly. In
the third column we simulated binding events occurring at a much slower
timescale, comparable to the frame time. In all cases, the shape of the
Biophysical Journal 96(2) 707–716
ccRICS in Live Cells709
RICS function is substantially different for these two processes: diffusion
and binding. Only when the binding (or blinking) is extremely fast (in the
microsecondtimescale), the RICSfunction due to fast diffusion and blinking
of immobile molecules tends to have similar shape, however, most biolog-
ical binding equilibria are not that fast.
Bleed through artifacts
The molecular information from the ccRICS can be compromised if there is
a large bleed through between the two channels. We measured the bleed
through in our microscope and filter combination using cells expressing
only one color; it was ~2% of the green emission into the red channel.
The bleed through of the red emission into the green channel was below
1%. As we discussed in the manuscript, the cross correlation amplitude
can be at most as large as the autocorrelation. A way to express the amount
of cross correlation is to normalize the amplitude of the cross correlation to
the amplitude of the autocorrelation, resulting in an index of cross correla-
tion between 0 and 1. The question is how can we distinguish true cross
correlation from bleed through?
According to the expression for the G(0,0) in Eq. 2, the index of the
ccRICS signal will be 1 due to bleed through if only one species is present
in the sample. Therefore, to properly interpret the cross correlation index, we
need to estimate the spectral bleed through and the amount of signal in the
two channels. In our cellular systems, we have some autofluorescence,
which is less 10% of the total fluorescence and approximately an equal
number of green and red molecules. If all molecules will be uncorrelated,
the presence of bleed through (2%) will give a cross correlation index of
0.04 according to Eq. 2. This estimation will be incorrect if we had much
less red molecules than green molecules. In Table 1 we report several
measurements with a cross correlation index below 0.1 and in some cases
below 0.05. When the index is below 0.05, we attribute this value to bleed
through.In general,it seemsthat thebleedthroughbetweenthe twochannels
does not impact our conclusions about the existence and quantification of
protein complexes because in many cases we observe cross correlation
indexes well above 0.05. Because the bleed through is small in our instru-
ment, the value of the cross correlation index (from 0 to 1) is approximately
equal to the fraction of correlated molecules so that the value of this index
can be directly related to the fraction of complexes containing both mole-
cules. Accordingly, we should be able to observe even 5% of molecules
carrying both colors.
Equations used for fitting the RICS function
The RICS correlation function in its simpler form can be written as
the product of two terms. One term corresponds to the effect of
diffusion and how the intensity at one pixel propagates to the next
neighbor pixel. This term is similar to the normal time dependent term
in fluctuation spectroscopy but it accounts for the difference in time
between the horizontal line and the vertical line in the raster scan data
surface with a diffusion constant of 5 mm2/s. Ten particles, carrying both colors are diffusing in the same area with the same diffusion constant.The upper panel
shows the RICS function obtained from analyzing the data at one channel and the second panel shows the ccRICS. The intensity of the green particles bleeds
into the red channel (2%) to simulate bleed through. Under this condition, in which there are red particles diffusing, the bleed through causes a very small
difference on the amplitude of the ccRICS. Column 2: particles bind and unbind rapidly (with respect to the line time) to and from fixed locations. The
RICS (autocorrelation) or ccRICS have the same shape. This shape is quite different from the shape obtained with particles diffusing. In the x direction,
the shape is related to the extension of the illumination volume. Column 3: particles undergo slow (with respect to the line time) binding equilibria. The shape
of the RICS (or ccRICS) is related to the shape of the illumination volume.
Simulationof particlesdiffusingandbindingto fixedlocations. Column1:simulationof100particles oftwodifferentcolorsdiffusingina square
Biophysical Journal 96(2) 707–716
710 Digman et al.
1 þ4Dðtpx þ tljÞ
1 þ4Dðtpx þ tljÞ
In this equation, D is the diffusion coefficient in units of mm2/s, tpand tlare
the pixel dwell time and the line time in s, respectively and w0is the waist
(1/e2) of the PSF in microns. g is a factor that account for the profile of illu-
mination (0.35 for 3D Gaussian and 0.076 for Gaussian Lorentzian, respec-
tively) and N is the number of molecules in the excitation volume. The
second term of the RICS autocorrelation function reflects the apparent
broadening of the PSF due to the diffusion of molecules. In the absence
of diffusion, this term is just the spatial correlation of the PSF, which we
describe with a Gaussian. When diffusion is present, the width of this
Gaussian term becomes time dependent as shown below:
Sðx;jÞ ¼ exp
In this expression, dr is the pixel size, in microns.
The overall RICS correlation function is given by the product of the two
Fig. 2 shows the images of a mouse embryo fibroblast,
expressing vinculin-EGF (green channel) and paxillin-
mCherry (red channel) whose fluorescence is captured in
the two different channels. These two proteins colocalize
well as shown by comparing the images in the two channels
and their RGB overlay (Fig. 2, A–C). FAK-EGFP and
paxillin-mCherry colocalize similarly (Fig. 3). However,
we were unable to observe FRET using either intensity
methods or FLIM (data not shown). Thus despite the coloc-
alization and their coimmunopurification, there is no direct
evidence suggesting that they interact in cells. In the Sup-
porting Material, we show movies of the image stacks
used for Figs. 2 and 3 for the green and red channels, respec-
tively. These movies show the apparent movement of the
In contrast, both pairs (vinculin-EGFP and paxillin-
mCherry and FAK-EGFP and paxillin-mCherry) cross-
correlated, which shows that they interact in cells. Figs. 2
and 3 show the RICS autocorrelation and cross correlation
functions for vinculin-paxillin and FAK-paxillin, respec-
tively. We used two different moving average lengths for
the removal of the immobile fraction. The first row of
RICS functions (Figs. 2 and 3, D–F) were obtained with
a high pass filter (moving average of 10 frames) that corre-
sponds to ~11.5 s; i.e., all processes from microseconds to
~11.5 s are present in the RICS function. The presence of
RICS autocorrelation for each of the individual channels
shows that both vinculin and paxillin are diffusing relatively
rapidly in the cytoplasm. The characteristic elongated shape
of the RICS function along the fast scan axis (Figs. 2 and 3)
shows that the molecules are moving fast relative to the line
scanning time, which is ~4.325 ms. Table 1 reports the
results from fitting the RICS function using a one-species
diffusion model (8) for several cells.
The amplitude of the ccRICS function (Fig. 2 F), using the
10-frame moving average, is much less than that for the auto-
correlation (Fig. 2 D). This small ccRICS signal likely corre-
sponds to a small amount of spectral bleed through.
However, when the moving average is set to 40 frames
(Fig. 2 I), the amplitude of the ccRICS increases
VIN-PAX moving average of 10
VIN-PAX moving average of 40
FAK-PAX moving average of 10
FAK-PAX moving average of 40
G is the G(0,0) term and D is the apparent diffusion coefficient. Indexes 1, 2, and cc refer to channel 1, channel 2, and the cross correlation between the two
*Indicates quiescent regions of the cell.
Biophysical Journal 96(2) 707–716
ccRICS in Live Cells 711
substantially showing molecular interaction. In Table 1 we
report the ratios between the G(0) obtained for the ccRICS
and the G(0) obtained for the average RICS signal (autocor-
relation) in the two channels. This ratio is relatively small
(<0.1) when the moving average length is set at 10 frames
but increases to ~0.4 when the moving average length is
set to 40 frames (Fig. 2 I). For the data analyzed with the
moving average of 40, the shape of the ccRICS function is
round, rather than elongated, and it has the size of the point
spread function (PSF). This shows that the correlated move-
ments of the two proteins at this (slower) timescale are due to
localized binding-unbinding equilibria rather than diffusion.
sity, the overall macroscopic apparent motion of the entire
adhesion starts to show in the shape of the RICS function.
The ccRICS analysis was repeated for the pair FAK-EGFP
and paxillin-mCherry (Fig. 3). Again, we found individual
molecules diffusing rapidly in the cytoplasm but there was
no cross correlation using a moving average of 10 frames
(Fig. 3 F). This shows that the two molecular species are
diffusing rapidly, but independently. However, when we
increased the length of the moving average operation to
emphasize slower fluctuations, the ccRICS signal became
significant (Fig. 3 I). The shape of the RICS function for
the slow process was no longer elongated and approximated
the size of the point spread function, showing that the cross
correlated events correspond to slow binding-unbinding
equilibria, as described above. Interestingly, the ratio of the
amplitudes of the ccRICS to the RICS autocorrelation func-
tions for this pair in this particular cell is smaller than that for
the vinculin-paxillin pair.
Although the above analyses were obtained by averaging
the spatial correlations over the entire frame, the RICS anal-
ysis also can show interactions in a smaller region(s) of
interest (ROI), thus providing a map of where the protein
interactions occur within the cell. To show any local differ-
ences in the interactions, we systematically calculated the
ccRICS in a small ROI (64 ? 64 pixels) scanned across
the entire image. We normalized the ccRICS by dividing
the G(0,0) of the cross correlation signal by the average
of the autocorrelation function for the two channels in the
same ROI. We found that the ccRICS signal is higher in
the regions where the focal adhesions are disassembling,
e.g., at the upper and right border of the image in Fig. 4 A.
Channel greenChannel redCross-correla?on
green and red channels. The size of the image is 12.5 mm by 12.5 mm. (D–F) RICS auto (channel 1 and 2) and cross correlation (ccRICS) signal using a moving
average of 10 frames to remove the quasi-immobile components. The RICS function is the representation of the autocorrelation function described by Eq. 1
(see text). The axes in the plane represents the x and j increments and the vertical axis is the G(x,j) function. (G–I) RICS functions obtained using a moving
average of 40 frames. The fits of the RICS data using one diffusion component are shown in Table 1.
(A–C) Intensity images of a cell expressing vinculin-EGFP and paxillin-mCherry in the green and red channels and the RGB composition of the
Biophysical Journal 96(2) 707–716
712Digman et al.
We interpreted this cross correlation as due to complexes
containing both proteins that are being released from adhe-
sions during their disassembly. Little or no cross correlation
is seen away from disassembling adhesions, suggesting that
the complexes, after detaching from the adhesions, have only
a brief life, and fall apart quickly. It also suggests that there
are few if any preassembled complexes in the general cyto-
plasm or associated with the other adhesions, on this time-
scale. Fig. 4 B uses a moving average of 40 frames. This
analysis emphasizes the locations in the cell where binding
equilibria are more prominent. The map of these interactions
has large amplitude in the upper-right part of the image,
where adhesions are disassembling and ‘‘sliding’’ toward
the lower left corner of the image. This suggests that
complexes containing both molecules release from adhe-
sions that are disassembling and/or sliding.
Cross correlation of a mutant of FAK that does not
bind to Paxillin
As a control we present the RICS and ccRICS analysis of
a cell expressing a FAK mutant that is not supposed to
bind to paxillin (Fig. 5). For this cell, the FAK mutant
labeled with EGFP does not concentrate at the adhesions
whereas paxillin-mCherry is shown at the adhesion. The
cross correlation is very small and it can be attributed to
the bleed trough effect.
Techniques developed to achieve super-resolution like pho-
toactivated localization microscopy and stochastic optical
reconstruction microscopy are basedon intensityfluctuations
of single molecules induced by external means (1,14). The
RICS method is also based on fluctuations, but they are
number fluctuations due to the presence of a molecule or
complex in the excitation volume. In the super-resolution
techniques, the induced fluctuation must be maintained
long enough to determine the position of the particle. During
the fluctuation (or sampling time) the particle cannot move.
Thus, there is a trade-off between the duration of the fluctua-
tion and the spatial resolution. In the RICS method intensity
fluctuations are measured on a very fast timescale. The
Channel greenChannel red Cross correla?on
green and red channels. The size of the image is 12.5 mm by 12.5 mm. (D–F) RICS auto (channel 1 and 2) and cross correlation signal using a moving average
of 10 frames to remove the quasi-immobile components. (G–I) RICS functions obtained using a moving average of 40 frames. The fits of the RICS data using
one diffusion component are shown in Table 1.
(A–C) Intensity images of a cell expressing FAK-EGFP and paxillin-mCherry in the green and red channels and the RGB composition of the
Biophysical Journal 96(2) 707–716
ccRICS in Live Cells 713
change in position of the particle during the sampling time is
used to determine its physical properties such as mobility,
composition and number. However, this information content
comes at the expense of spatial resolution. The RICS method
is intrinsically 3D and relatively low resolution. When
compared with the single molecule imaging methods, the
length of 10 frames.(B) Obtained usinga moving averageof 40 frames.The upper part of the cell correspondsto the region where adhesionsare disassembling.
Map of the ccRICS signal normalized to the average of the autocorrelation function in both channels. (A) Obtained using a moving average
Channel green Channel redCross-correla?on
and is not concentrated at the adhesion. (B) Paxillin binds to the adhesion. (C and D) RICS autocorrelation of the green and red channels, respectively after
immobile subtraction using a moving average length of 40. (E) The ccRICS is virtually nil except for the bleed through effect.
Cell expressing FAKmut-EGFP and PAX-mCherry. The size of the image is 12.5 mm by 12.5 mm. (A) The FAK mutant is not binding to Paxillin
Biophysical Journal 96(2) 707–716
714Digman et al.
fluctuation analysis can better characterize the nature of the
particle and the physical origin of the intensity fluctuation.
Although we use the temporal part of the intensity fluctua-
tions to determine mobility and complex composition in
this study,it ispossibletousetheamplitudeof thefluctuation
to determine the stoichiometry of the fluctuating particles
We have applied the RICS method to study the dynamics
of cell-substrate adhesion complexes. These adhesions can
be highly dynamic, and form and disassemble in minutes
or less. Over 100 proteins (15) associate with adhesions,
most of which are thought to reside in multimolecular
complexes, based on immunoprecipitation and/or colocaliza-
tion studies. A critical biological question, therefore, is
which proteins do in fact reside in such a discrete structural
complex and in what ratio, and when and where do these
interactions occur. In this context, we have asked whether
adhesions form or disassemble by the addition or subtraction
of preassembled complexes.
We studied two pairs of proteins that are known to interact
by coimmunoprecipitation and colocalize in adhesions (16).
Surprisingly, we found that neither of the protein pairs are in-
teracting in the cytoplasm, suggesting that adhesions do not
cytoplasmic complexes. However, we do see complexes at
disassembling adhesions. In these cells, the cytoplasmic
concentration of these molecules is low—probably too low
to support their spontaneous association by mass action.
From this, a picture emerges in which the assembling adhe-
sion itself serves as a scaffold for the assembly of the protein
complexes. In contrast, the large adhesions that we have
studied disassemble by the release of complexes. In disas-
sembling edges of adhesions, we see FAK-pax and vincu-
lin-pax complexes exchanging from/leaving adhesions, and
their subsequent diffusion away from the adhesion.
However, the complexes are short lived and are not seen
far from the adhesion, suggesting that they disassemble by
mass action once out of the adhesion. As a consequence of
the monomer addition (assembly) and large aggregate
subtraction (disassembly), the adhesion appears to move or
slide gradually. Thus, this sliding movement seems to be
due to a treadmilling rather than the macroscopic movement
(sliding) of the entire adhesion (17,18). This distinction was
shown elegantly in a recent study of the apparent movement
of adhesions using the photoactivated localization micros-
copy technique (19). It is important, therefore, to distinguish
localized molecular events from macroscopic intensity
changes, which are the sum of many events. These observa-
tions were made possible by large dynamic range of ccRICS,
which simultaneously measures the diffusion, composition,
and exchange of complexes.
The ccRICS methodology provides rich information about
protein interactions. It can distinguish between different
phenomena such as diffusion of protein aggregates and
binding events involving protein complexes. We were able
to separate these events on the basis of timescale and the
different RICS functions that these two events provide. To
illustrate the different signatures of diffusion and binding,
we carried out simulations of particles diffusing in a plane
(Fig. 1). We found that binding and diffusion occur at two
different timescales, and we were able to separately identify
the two processes and to quantify the timescale and the level
of cross correlation on the basis of the different shape and
amplitude of the ccRICS function (elongated versus round).
The spatial map of the ccRICS amplitude provides further
information about the cell compartments (adhesion regions)
where some interactions occur preferentially. The spatial
information shows that there is a gradient of protein
complexes that are emanating from the disassembling adhe-
sions. The existence of this gradient shows that we cannot
calculate the equilibrium constant of the complex formation
from the relative population of the free diffusing molecules
and complexes. The formation of the complex is catalyzed
by the scaffold at the adhesion and not by the mutual affinity
of the individual proteins. It is the specific spatial informa-
tion provided by ccRICS that allow us to establish the nature
of the interactions and to properly interpret the origin of the
The model of the disassembling of focal adhesions arising
from the ccRICS implies that relatively large protein aggre-
gates detach from the adhesions and then rapidly crumble in
small parts that are still visible at a short distance from the
disassembling adhesions. One important question is the stoi-
chiometry of these aggregates immediately after they detach
from the adhesion, because they could show the size and
composition of the complexes at the adhesions. From the
ccRICS studies we can only infer that the aggregates contain
multiple proteins. In a previous study we determined that
there are ~8–10 copies of paxillin in these disassembling
aggregates (18). From the existence of a cross correlated
signal and from the ratio of the G(0,0) for the cross correla-
tion to the G(0,0) of the autocorrelation we infer that both
vinculin and FAK are also present in multiple copies. A
more quantitative analysis will require the statistical analysis
of the fluctuation amplitudes and their correlations in the two
In conclusion, RICS and ccRICS carry the information
about the existence, composition, and dynamics of molecular
complexes. Colocalization methods are insufficient to estab-
lish molecular aggregation, and coimmunoprecipitation does
not establish that a particular interaction occurs in cells. By
exploiting correlated molecular fluctuations available in
single, dual channel confocal images, we can unequivocally
establish the formation of molecular complexes in live cells
and map their location. In addition, we can also separate
binding from diffusion and eliminate problems due to
bleaching, immobile components, and movement of edges
and other structures. In contrast to previous image based
two-color correlation measurements, which have two photon
Biophysical Journal 96(2) 707–716
ccRICS in Live Cells 715
excitation or other sophisticated hardware, two-color
ccRICS can be done on commercial scanning confocal
microscopes, and therefore is accessible to many more
researchers in the health and biomedical sciences.
Two movies are available at http://www.biophysj.org/biophysj/supplemental/
We thank Jenny Sasaki for cultivating and transfecting the MEF cells.
This work was supported in part by U54 GM064346 Cell Migration Consor-
tium (M.D., A.H., E.G.) National Institutes of Health grants P41-RRO3155
and P50-GM076516 (E.G.), the Natural Sciences and Engineering Research
Council of Canada, and the Canadian Institutes of Health Research
1. Betzig, E., G. H. Patterson, R. Sougrat, O. W. Lindwasser, S. Olenych,
et al. 2006. Imaging intracellular fluorescent proteins at nanometer
resolution. Science. 313:1642–1645.
2. Gustafsson, M. G. 2008. Super-resolution light microscopy goes live.
Nat. Methods. 5:385–387.
3. Weiss, S. 2000. Shattering the diffraction limit of light: a revolution in
fluorescence microscopy? Proc. Natl. Acad. Sci. USA. 97:8747–8749.
4. Willig, K. I., S. O. Rizzoli, V. Westphal, R. Jahn, and S. W. Hell. 2006.
STED microscopy reveals that synaptotagmin remains clustered after
synaptic vesicle exocytosis. Nature. 440:935–939.
5. Jares-Erijman, E. A., and T. M. Jovin. 2006. Imaging molecular inter-
actions in living cells by FRET microscopy. Curr. Opin. Chem. Biol.
6. Liebler, D. C. 2002. Proteomic approaches to characterize protein modi-
fications: new tools to study the effects of environmental exposures.
Environ. Health Perspect. 110 (Suppl 1):3–9.
7. Bacia, K., S. A. Kim, and P. Schwille. 2006. Fluorescence cross-corre-
lation spectroscopy in living cells. Nat. Methods. 3:83–89.
8. Digman, M. A., C. M. Brown, P. Sengupta, P. W. Wiseman, A. R. Hor-
witz, et al. 2005. Measuring fast dynamics in solutions and cells with
a laser scanning microscope. Biophys. J. 89:1317–1327.
et al. 2005. Fluctuation correlation spectroscopy with a laser-scanning
microscope: exploiting the hidden time structure. Biophys. J. 88:
10. Brown, C. M., R. B. Dalal, B. Hebert, M. A. Digman, A. R. Horwitz,
et al. 2008. Raster image correlation spectroscopy (RICS) for measuring
fast protein dynamics and concentrations with a commercial laser scan-
ning confocal microscope. J. Microsc. 229:78–91.
11. Wiseman, P. W., J. A. Squier, M. H. Ellisman, and K. R. Wilson. 2000.
Two-photon image correlation spectroscopy and image cross-correla-
tion spectroscopy. J. Microsc. 200:14–25.
12. Das, R., S. Hammond, D. Holowka, and B. Baird. 2008. Real-time
cross-correlation image analysis of early events in IgE receptor
signaling. Biophys. J. 94:4996–5008.
13. Petersen, N. O., and E. L. Elson. 1986. Measurements of diffusion and
chemical kinetics by fluorescence photobleaching recovery and fluores-
cence correlation spectroscopy. Methods Enzymol. 130:454–484.
14. Rust, M. J., M. Bates, and X. Zhuang. 2006. Sub-diffraction-limit
imaging by stochastic optical reconstruction microscopy (STORM).
Nat. Methods. 3:793–795.
15. Zaidel-Bar, R., S. Itzkovitz, A. Ma’ayan, R. Iyengar, and B. Geiger.
2007. Functional atlas of the integrin adhesome. Nat. Cell Biol.
16. Choi, C., M. Vicente-Manzanares, J. Zareno, L. A. Whitmore, A. Mo-
gilner, et al. 2008. Actin and alpha-actinin orchestrate the assembly and
maturation of nascent adhesions in a myosin II motor-independent
manner. Nat. Cell Biol. 10:1039-1050.
17. Wehrle-Haller, B., and B. A. Imhof. 2003. Actin, microtubules and
focal adhesion dynamics during cell migration. Int. J. Biochem. Cell
18. Digman, M. A., C. M. Brown, A. R. Horwitz, W. W. Mantulin, and E.
Gratton. 2008. Paxillin dynamics measured during adhesion assembly
and disassembly by correlation spectroscopy. Biophys. J. 94:2819–
19. Shroff, H., C. G. Galbraith, J. A. Galbraith, and E. Betzig. 2008. Live-
cell photoactivated localization microscopy of nanoscale adhesion
dynamics. Nat. Methods. 5:417–423.
716 Digman et al.
Biophysical Journal 96(2) 707–716