© 2008 Nature Publishing Group http://www.nature.com/natureneuroscience
TECHNICAL REPORT
Three-dimensional random access multiphoton
microscopy for functional imaging of neuronal activity
Gaddum Duemani Reddy1,2, Keith Kelleher3, Rudy Fink2 & Peter Saggau1,2
The dynamic ability of neuronal dendrites to shape and
integrate synaptic responses is the hallmark of information
processing in the brain. Effectively studying this phenomenon
requires concurrent measurements at multiple sites on live
neurons. Substantial progress has been made by optical
imaging systems that combine confocal and multiphoton
microscopy with inertia-free laser scanning. However, all
of the systems developed so far restrict fast imaging to two
dimensions. This severely limits the extent to which neurons
can be studied, as they represent complex three-dimensional
structures. Here we present a new imaging system that utilizes
a unique arrangement of acousto-optic deflectors to steer a
focused, ultra-fast laser beam to arbitrary locations in
three-dimensional space without moving the objective lens.
As we demonstrate, this highly versatile random-access
multiphoton microscope supports functional imaging of
complex three-dimensional cellular structures such as
neuronal dendrites or neural populations at acquisition
rates on the order of tens of kilohertz.
Developing a thorough understanding of the computational capabilities of a single neuron is one of the primary goals in the field of
experimental neuroscience. However, any attempt to study neuronal
physiology with subcellular resolution is complicated by four fundamental technical challenges. First, the small size of many neuronal
processes prohibit direct electrical recording. For hippocampal CA1
pyramidal neurons in particular, an estimated 50% of all synapses are
located on thin dendrites that are less than 1 mm in size1 and are thus
inaccessible to patch pipettes. Second, in brain slices or intact cortex,
neurons of interest can be located several hundred microns deep inside
the tissue, requiring that the recording method be capable of isolating
them from their environment. Third, the relevant neuronal signals,
such as transient membrane potentials and ionic membrane currents,
happen on the order of milliseconds, necessitating recording methods
with high temporal resolution. Finally, the space occupied by these cells
can span hundreds of microns in all three dimensions2, further
restricting optimally effective recording methods to only those capable
of monitoring such a volume. These four challenges have stimulated
the development of increasingly advanced fluorescence imaging
techniques, which have become commonly valued tools in the
arsenal of the experimental neuroscientist.
Indeed, several studies have met the first two of the challenges listed
above by using the diffraction-limited spatial resolution and opticalsectioning capabilities provided by confocal and multiphoton microscopy to study physiological dynamics in thin oblique dendrites3–6 and
dendritic spines7–13. Recent studies have further extended the temporal
resolution of these techniques by replacing galvanometer-driven beamsteering mirrors with inertia-free acousto-optic deflectors (AODs)14–18.
The combination of multiphoton/confocal microscopy with this form
of beam scanning15,16 represents the only approach that truly meets the
first three of the challenges listed above. This is because the random
access capabilities and high temporal resolution provided by AODbased scanning allows for effectively simultaneous imaging of several
sites on a neuron19, whereas similarly equipped galvanometer-based
scanning methods are limited to nearby locations and line-scans.
To date, however, no technique has effectively met all four of the
challenges, primarily because it is only possible to image from a single
focal plane with sufficient temporal resolution to track neuronal
signals. Indeed, even for AOD-based scanning techniques, the temporal
resolution enhancement and random-access capabilities apply only to a
single focal plane. In such techniques, scanning different axial planes
requires the objective lens to be physically moved with an actuator, such
as a stepper motor or a nanopositioner, a process that is temporally
limited by the combined inertia of objective lens and actuator. Several
methods have been developed to quickly change the focal position to
allow for axial scanning, including deformable mirrors20, variablefocus liquid-filled lenses21 and simultaneous spatial and temporal
focusing22,23. However, none of these techniques have the ability to
simultaneously scan in the lateral dimension, and none have been
demonstrated to functionally image parameters of neuronal activity. In
addition, very few methods have achieved acquisition rates faster than
1 kHz. Other techniques that extend the depth of field to provide a
maximum projection image of the neuron, such as axicon-based
scanning24,25, prevent the functional imaging of structures that overlap
axially and markedly increase the power requirements for multiphoton
imaging. In addition, a recent technique of strategically combining
axial movement of the objective lens via a piezoelectric actuator with
lateral scanning using galvanometers has been shown to be effective
1Department of Bioengineering, Rice University, 6100 Main Street, Suite 116 Keck Hall, Houston, Texas 77005, USA. 2Department of Neuroscience, Baylor College of
Medicine, One Baylor Plaza S730, Houston, Texas 77030, USA. 3Department of Biology and Biochemistry, University of Houston, 4800 Calhoun Road, Houston, Texas
77004, USA. Correspondence should be addressed to P.S. ([email protected]).
Received 17 December 2007; accepted 1 April 2008; published online 27 April 2008; doi:10.1038/nn.2116
NATURE NEUROSCIENCE VOLUME 11
[
NUMBER 6
[
JUNE 2008
713
TECHNICAL REPORT
Figure 1 Two-dimensional versus threef2
dimensional AOD Scanning. (a) Two-dimensional
FAOL
AOD
scanning. Left, angular deflection (y(x,t)) of a
AOD1
collimated laser beam by a single AOD with
(x,t )
(x,t )
a fixed frequency (f ) acoustic wave. Right,
AOD2
two-point random-access lateral scan of an ultraf
fast laser beam focused by an 10 objective lens
f1
into a cuvette containing fluorescein solution,
resulting in two distinguishable focal spots in a
single axial plane (inset, zoomed-in view).
Bottom, fast scanning with two orthogonal
AODs was restricted to a single plane. (b) Threedimensional scanning. Left, change in collimation
(focal length FAOL) and angular deflection (y(x,t))
by counter-propagating chirped acoustic waves
(f1, f2) in a pair of AODs. Right, two-point random-access axio-lateral scan, resulting in two distinguishable focal spots at different axial planes
(inset, zoomed-in view). Bottom, two orthogonal pairs of AODs supported fast scanning in a volume.
b
in imaging somatic calcium transients in vivo26. However, although this
technique employs sinusoidal axial movements, it is still constrained
by the inertia of the objective lens, which limits the acquisition rate
of the system.
Given the extent of these various constraints in imaging highly
complex structures such as neuronal dendrites, we recently developed a
strategy that utilizes a unique arrangement of multiple AODs to allow
for inertia-free random-access scanning in three dimensions27. In this
proof-of-principle study, we demonstrate that this technique allows for
scanning acquisition rates on the order of tens of kilohertz for userselected sites of interest that occupy many different positions in a
volume. In addition, we show how this scheme has the inherent benefit
of compensating for spatial dispersion, an effect that degrades the
spatial resolution of AOD-based multiphoton microscopy and that
usually requires an external source for compensation28,29. Another
recent report has used a simpler version of this approach for fast
scanning of neurons in three dimensions. However, this reduced
method does not allow for true random-access imaging in all three
dimensions and therefore has not demonstrated fast three-dimensional
functional imaging30.
In the current study, we implemented our three-dimensional scanning technique and combined it with multiphoton microscopy to
Three-dimensional
AOD scanner
Ti:S laser
(Mira HP)
RESULTS
Three-dimensional acousto-optic scanning principle
To generate random access scans in three dimensions, we used a system
of AODs with chirped (that is, continuously changing in frequency)
and counter-propagating acoustic waves. This approach is different
from the method used for two-dimensional random-access applications14–17,31–33, in which a fixed frequency is supplied to a single AOD
per scan dimension (Fig. 1a), resulting in a deterministic deflection
angle Y, as specified by equation (1),
lf
ð1Þ
v
where l is the wavelength of the incident light, and f and v are
the frequencies and velocities of the propagating acoustic waves,
respectively34,35. As detailed in an earlier work27, when two AODs are
y¼
b
Scan
mag.
(1.2:1)
Short-pass dichroic mirror
Autocorrelator
Emission
filter
Spectrum
analyzer
Long-pass dichroic mirror
PMT
λ /2 plate
create the first system capable of overcoming all four of the limitations
listed above and found the first evidence of random-access functional
imaging of neurons in a volume of tissue. Furthermore, we used this
system to concurrently monitor calcium transients at several different
lateral and axial locations on individual pyramidal neurons in
brain slices.
Relay
(1:1)
Relay
(1:1)
X2 scan
Y2 scan
Demag
(1:5.6)
Objective
Axial distance from objective focus (µm)
Beam
expander
a
Re
(1 lay
:1
)
© 2008 Nature Publishing Group http://www.nature.com/natureneuroscience
a
30
Predicted
20
Measured
10
0
–10
–20
–30
–7.6
–5.7
–3.8
–1.9
0
1.9
3.8
5.7
7.6
Chirp parameter (Mhz µs–1).
Y1 scan
X1 scan
Figure 2 Optical layout and axial scan range. (a) Imaging system containing an ultra-fast pulsed laser (titanium:sapphire, Ti:S Mira HP), pulse diagnostic
devices (spectrum analyzer and autocorrelator), beam-shaping optics (beam expander and scan magnification), the three-dimensional AOD scanner (see inset),
an upright epifluorescence microscope (dichroic mirrors and objective lens) and the detection unit (demagnification, emission filter, PMT). Inset, threedimensional AOD scanner consisting of polarization control (l/2 plate) and four AODs with relay telescopes (scan X1, X2, Y1 and Y2). An infrared-sensitive
camera was located above the short pass dichroic mirror, and used for trans-illumination visualization of the specimen during patching (not shown).
(b) Predicted (solid line) and measured (crosses) axial distance of the focal position from the inherent focal plane of the objective lens.
714
VOLUME 11
[
NUMBER 6
[
JUNE 2008 NATURE NEUROSCIENCE
TECHNICAL REPORT
a
c
2+1D
Axial
Lateral
Axial
Lateral
Axial
© 2008 Nature Publishing Group http://www.nature.com/natureneuroscience
b
3+0D
+25.0 µm
Lateral:1.59
Axial:1.49
+12.5 µm
Lateral:1.08
Axial:1.13
Objective
focal
plane
Lateral:1.00
Axial:1.00
–12.5 µm
Lateral:0.92
Axial:1.06
–25.0 µm
Lateral:1.36
Axial:1.13
Figure 3 Structural imaging with different three-dimensional scanning
modes. (a) Axial and lateral projections of a fluorescent bead (10 mm in
diameter) that was imaged with combined two-dimensional AOD lateral
scanning and stepper-motor axial positioning (2+1D mode), and with threedimensional AOD scanning (3+0D mode). The resolution of both scans was
250 250 100 points. (b) Axial projections (left) and selected optical
sections (right) of a reconstructed pollen grain (B35 mm) imaged in both the
2+1D and the 3+0D mode (scale bars, 10 mm). Resolutions of both scans
were 250 250 100 points. (c) Lateral and axial point PSFs at different
depths (left), as determined by imaging 200-nm fluorescent beads. Relative
changes (right) in the full-width half maximum of the PSFs when compared
with inherent objective focus (scale bars, 1 mm).
2+1D
3+0D
used per lateral
counter-propagating
independent effects
lateral deflection of
equation (2),
dimension and the acoustic waves are
and chirped (as shown in Fig. 1b), two
are observed. First, there is a deterministic
the laser beam with an angle specified by
lðf1c f2c Þ
ð2Þ
v
where f1c and f2c are the center acoustic frequencies in the first and
second deflectors, respectively. Second, a beam convergence or divergence is introduced as an effective cylindrical lens is generated with a
focal length, FAOL, as described by equation (3),
y¼
Three-dimensional random access multiphoton microscope design
Using the principle of three-dimensional AOD-based scanning
described above, we constructed a multiphoton microscope system
(Fig. 2a). The laser source was provided by a titanium:sapphire laser
(Mira HP, Coherent) pumped by a 532-nm solid-state laser (Verdi 18,
Coherent) to generate ultra-fast pulses around 800 nm with a modelocked average power of approximately 4 W (Dl, B10 nm; pulse width,
B200 fs). The B3-mm output beam was then magnified threefold and
relayed through the scanner, which consisted of a series of four largeaperture AODs, four 1:1 telescopes and a 1.2:1 scan-angle magnifier,
before reaching the back focal aperture (BFA) of a 60 (NA, 1.0) waterimmersion objective lens (CFI Fluor, Nikon). The scan-angle magnification served to increase the effective lateral scan field to B200 mm 200 mm. The laser scanner was coupled to the framework of an upright
epifluorescence microscope (E600FN, Nikon), in a manner similar to
that of previously developed systems15.
a
2
v
ð3Þ
2la
where a, expressed in Mhz ms–1, is the chirp parameter, or the rate of
change in frequency27,36.
By using two orthogonal pairs of deflectors, for a total of four AODs,
we created a variable focal–length spherical lens that can be used to
slightly change the collimation of a beam entering the objective lens,
thus changing the axial focal positions. Because the deflection angle, y,
depends only on the difference between the center acoustic frequencies
in each pair of AODs, whereas the focal length of the acousto-optic–
based lens, FAOL, depends only on the rate of change in the acoustic
frequencies, this approach allows for independent axial and lateral focal
positioning. This feature, when combined with multiphoton excitation
principles, results in the ability to precisely measure fluorescence from
anywhere in a volume.
FAOL ¼
b
c
Beam converging
z
Objective moved
down 25 µm to
compensate
Reference plane
AOD focus : 25 µm
above focal plane
y
x
Figure 4 Structural imaging of neurons. (a) An axial projection of a neuron
filled with a fluorescent label (Alexa 594) is shown, acquired in three
separate image stacks using the 3+0D scanning mode (axial range, 50 mm;
scale bar, 50 mm). The resolution of the scans was 200 200 20 points.
(b) Full projections of single image stacks from another neuron, acquired
using both the 3+0D and the 2+1D mode (xy scale bar, 50 mm; zy scale
bar, 10 mm). The resolution of both scans was 200 200 50 points.
(c) Dendritic segment imaged in 3+0D mode with high resolution (zoom-in)
at three different AOD-controlled focal distances. Middle, reference image,
taken at inherent focal distance of the objective lens (reference plane). Test
images, taken at AOD-controlled effective focal planes 25 mm above (top)
and 25 mm below (bottom) the reference plane with the objective lens moved
in the opposite direction to compensate (scale bars, 1 mm).
NATURE NEUROSCIENCE VOLUME 11
[
NUMBER 6
[
JUNE 2008
Beam collimated
3+0D
Reference plane
AOD focus : At
objective focal plane
z
Beam diverging
Objective moved up
25 µm to
compensate
Reference plane
y
x
2+1D
AOD focus : 25 µm
below focal plane
715
TECHNICAL REPORT
a
© 2008 Nature Publishing Group http://www.nature.com/natureneuroscience
–11 µm
1
–6 µm
2
2
Axial distance
from objective
lens focal plane
– 11 µm
– 6 µm
– 3 µm
–1 µm
1
–3 µm
b
–1 µm
1
2
100% ∆F/F
–50 mV
100 ms
Figure 5 Functional imaging of CA1 pyramidal neurons. (a) Functional
imaging sites selected from a structural image acquired in 3+0D scanning
mode. Left, individual optical sections (scale bars, 50 mm). Right, userselected sites on a maximum projection image (scale bar, 50 mm). Axial
distances from the inherent focal plane of the objective are color coded. The
resolution of the structural scan was 200 200 20 points. (b) Randomaccess imaging of calcium signals at the selected sites (top) during a train of
bAPs (bottom). Acquisition rate, B10 kHz.
We used large optical-aperture (9 mm 9 mm), slow-shear acoustic
wave TeO2 AODs (OAD-1121, Isomet). This material was chosen
because of its high acousto-optic figure of merit, which allows for
large-aperture devices with high interaction bandwidth (B40 Mhz)
and a high diffraction efficiency (B70% into the first order). In turn,
the large size of the apertures increases the resolution of the AODs, as
determined by the number of resolvable points, by reducing the
diffraction-limited spot size35. One consequence of using this
acousto-optic material, however, is that each deflector rotates the
polarization of the beam by 90 degrees34. This requires that the
AODs be arranged in a specific order to match their incident polarization, with orthogonally oriented deflectors placed in sequence. The
polarization for the first deflector was adjusted using a half-wave plate
(AHWP05M-950, Thorlabs).
Each AOD is driven by a radio-frequency signal generated by a
voltage-controlled oscillator (VCO, VCO-200A, Isomet). These particular VCOs have high tuning linearity (o0.1% deviation over
a 50-Mhz range) and very high tuning-slew rate (410 Mhz ms–1),
which exceeds the maximum chirp parameter that is needed for
716
our AOD-based focusing (defined as the entire frequency range
of 40 Mhz over the aperture time of 10 ms, or 4 Mhz ms–1). To reach
the radio-frequency saturation power of approximately 2 W for the
deflectors, we amplified the output from the VCOs (IA-100-3 and
DA-104-2, Isomet).
The control voltages to the VCOs were supplied by the four analogoutput channels of a multifunction data acquisition card (NI-6259,
National Instruments). The high analog-output rate (1.25 MHz)
resulted in relatively smooth frequency ramps being generated by the
VCOs. The fluorescence signal is detected, current-to-voltage converted, amplified by a photomultiplier tube (PMT) module (H7712-13,
Hamamatsu) and custom designed electronics, and received by an
analog-input line from the data acquisition card. The process of beam
scanning and data acquisition is controlled by custom software
designed in the Dot Net 2005 platform (Microsoft). By synchronizing
the hardware tasks, the software allows the user to carry out a threedimensional raster scan to locate regions of interest and subsequently
select recording sites in those regions.
Axial position scan
We determined the extent to which measured axial position deviations
agreed with theoretical values on the basis of Gaussian optics analysis.
For this purpose, we used the scanner to image a layer of 200-nm
fluorescent beads (T-14792, Molecular Probes) that we selectively
positioned at different axial distances from the objective lens. Practically, this was done by acquiring image stacks of the bead layer using an
objective-lens stepper motor (Remote Focus Accessory, Nikon) around
axial locations determined by the AOD scanner.
The initial stack was imaged at the inherent focal position of the
objective lens by using unchirped acoustic waves in each AOD.
For this initial stack, the microscope was adjusted manually to bring
the bead layer into focus. Subsequently, the axial location of the
specified section being imaged was moved by changing the chirp
parameter in each AOD. This changed the collimation of the beam
entering the BFA of the objective lens and shifted the axial location
of the focus relative to its inherent focus. The objective lens was
then repositioned to bring the bead layer back into focus using the
stepper motor, and another stack was acquired. The axial offset of the
objective needed to refocus the bead layer, as well as the position of
maximal fluorescence in the stack, were used to determine the extent to
which the axial focus position had shifted. We measured axial distances
(Fig. 2b), which closely matched theoretical predictions based on an
ABCD matrix theory model for Gaussian beam propagation37
that simplifies the entire optical layout to a spherical lens with a focal
length determined by equation (3), a 1:1.2 scan-angle magnifier and
an objective lens.
Three-dimensional imaging of test preparations
To demonstrate the ability of the microscope system to acquire full
three-dimensional images without moving the objective lens, we used
two types of fluorescent test preparations: 10-mm beads (F-8836,
Molecular Probes) and B35-mm pollen grains (30-4264, Carolina
Biological Supply). Initial tests were done with beads to compare the
structural images attained with two separate methods. To obtain a
control measurement, we used the AODs only for lateral positioning
and the stepper motor for axial positioning. This three-dimensional
scanning method, which we define here as ‘2+1D’ scanning, resembles
previously used schemes for AOD-based multiphoton laser scanning
microscopy14,15,17. The second three-dimensional scanning method,
which we define as ‘3+0D’ scanning, used only the AODs for both axial
and lateral positioning.
VOLUME 11
[
NUMBER 6
[
JUNE 2008 NATURE NEUROSCIENCE
TECHNICAL REPORT
a
b
–11
–9
–8
–6
2
11
19
–5
3
5
13
18
23
25
© 2008 Nature Publishing Group http://www.nature.com/natureneuroscience
z
x
y
50% ∆F/F
250 ms
c
50% ∆F/F
250 ms
–2
0
2
4
7
9
13
Figure 6 Fast three-dimensional monitoring of dendritic calcium dynamics.
(a) Three-dimensional functional image (using a random-access scan) of
calcium transients along oblique dendrites during a train of three bAPs
(n ¼ 5; scale bar, 50 mm). Acquisition rate, B3 kHz. (b) Three-dimensional
functional image (using a random-access scan) of a portion of an apical
50% ∆F/F
dendrite (n ¼ 5; scale bar, 25 mm). Acquisition rate, B5 kHz. Inset,
three-dimensional reconstruction. Lines indicate sites that are not easily
250 ms
distinguishable on the maximum projection image because they are located
above a brighter dendritic segment. (c) Three-dimensional functional image (using a random-access scan) of a zoomed-in portion of an oblique dendrite
demonstrating calcium transients in the dendrite, as well as in a spine (circled; n ¼ 5; scale bar, 2 mm). Acquisition rate, B3 kHz. For all images, axial
distances of the user-selected sites are color coded (measured in microns relative to the inherent objective lens focus). All values are rounded to the
nearest micron.
Using the image stacks, the test preparations were reconstructed
(AMIRA, Template Graphics Software) for each method, and axial and
lateral projections were created. Our typical results indicated that there
were only minimal differences between the 2+1D and 3+0D scanning
methods for both the 10-mm beads and the larger pollen grains
(Fig. 3a,b). More quantitative assessment of the resolution changes
can be achieved by comparing the point spread functions (PSFs) at
different axial locations versus the inherent objective focus (Fig. 3c and
Supplementary Fig. 1 online).
Three-dimensional imaging of neurons
Before a meaningful random access scan can be performed, a structural
image must be generated from which points of interest can be selected.
It would be convenient to generate this image using the 3+0D scanning
method, as this would allow for exact registration between the
structural and functional images. Our results from the test preparations
suggest that 3+0D scanning would produce images with relatively the
same quality as the 2+1D method. To determine the degree to which
this holds true for fluorescently labeled cells, we verified that thin
dendrites were visible with 3+0D scanning (Fig. 4a) and compared
structural images of hippocampal CA1 pyramidal neurons obtained
with 3+0D and 2+1D scanning (Fig. 4b), which showed minimal
differences between the two techniques.
To allow for comparisons on a more detailed level, we then zoomedin on a segment of a thin dendritic branch by restricting the acoustic
frequency ranges that were applied to the deflectors (Fig. 4c). We
imaged this segment at three different AOD-based focal positions to
NATURE NEUROSCIENCE VOLUME 11
[
NUMBER 6
[
JUNE 2008
compare the image quality when the focal plane is adjusted in this
manner. First, the beam entering the BFA remained collimated to
acquire an image of the segment at the inherent focal plane of the
objective lens (Fig. 4c). Note that this represents the image quality that
would be obtained from a single optical section with the 2+1D method.
Subsequently, the collimation was changed using the AODs to position
the effective focus approximately 25 mm above the inherent focal plane,
and the objective lens was moved down by the same amount with the
stepper motor to bring the dendritic segment back into focus (Fig. 4c).
Finally, the collimation was changed to position the effective focus
25 mm below the inherent focal plane, and the objective lens was moved
up accordingly to compensate (Fig. 4c). Comparing these images
(Fig. 4c) shows that small processes, such as dendritic spines, remain
visible at all axial locations, despite a marginal reduction in image
quality. These results confirm that exact registration during imaging
can be achieved by using 3+0D scanning to generate the initial
structural image of the cell, on which subsequent functional imaging
sites can be selected.
To test the ability of the three-dimensional random-access multiphoton microscope to provide multisite measurements of neural
activity, we monitored localized calcium transients in response to
back-propagating action potentials (bAPs). To do this, we carried out
a 3+0D structural scan over an axial range of up to 50 mm. From the
acquired stack of structural images, different points along the main
apical dendrite (Fig. 5) and oblique dendrites (Fig. 6) were selected for
functional imaging. The functional scan was subsequently carried out
by visiting only these selected sites with a 15-ms positioning time and a
717
TECHNICAL REPORT
© 2008 Nature Publishing Group http://www.nature.com/natureneuroscience
5-ms dwell time. The effective acquisition rate, which is the inverse of
the sum of these two values times the number of points, ranged from 3
to 10 kHz for the experiments shown. Note that, unlike previous twodimensional AOD-based scanning microscopes, the points that were
selected for the functional scan were from different lateral and axial
locations. In addition, user-selected points could be distinguished even
if they are primarily only axially displaced from each other (Fig. 6c),
which would be not possible in three-dimensional scanning methods
that markedly extend the depth of field24,25.
DISCUSSION
In conclusion, we have developed a random-access multiphoton
microscope that allows for fast three-dimensional imaging without
physically moving scan mirrors or the objective lens. This is achieved by
a unique arrangement of four AODs for laser-beam steering. Furthermore, we have used this microscope for structural and functional
imaging of pyramidal cells. Specifically, we have recorded calcium
transients from different lateral and axial positions on a neuron with
acquisition rates that are in the kilohertz range by using the randomaccess capabilities of this inertia-free three-dimensional laser scanner.
One current limitation of our approach is that it is restricted to a
50-mm axial scan range when using a high-magnification objective lens.
Although this exact number is flexible and the axial range can be
stretched further, we noticed a substantial drop in excitation power at
further axial distances. This is primarily because of the birefringent
diffraction principle of the slow-shear TeO2 AODs that we used.
A hallmark of this diffractive effect is a substantially narrow incident
angular aperture. This constraint reduces the diffraction efficiency of
uncollimated incident laser beams, such as the beams at the third and
fourth AODs of our scanner, when the axial position moves away from
the inherent objective focus. Taking into account the acoustic bandwidth considerations alone, an axial scan range of approximately
200 mm and a lateral scan range of approximately 350 mm should be
feasable27. One possible method for avoiding this effect would be to use
deflectors whose diffraction is not birefringent-based; unfortunately,
with larger angular apertures, such AODs have substantially lower
overall diffraction efficiencies. One possibility would be to utilize
specialized wide angular–aperture deflectors. These deflectors have
been shown to provide high diffraction efficiencies, while maintaining
larger angular apertures38,39, thus making them optimal for AODbased focusing applications. We are currently working on an improved
scanning scheme with extended axial scan range, using custom-made
wide angular–aperture AODs.
In general, larger axial ranges would also reduce the effective spatial
resolution. Indeed, in addition to minor reductions in effective
objective numerical aperature27, there is also an anticipated increase
in spherical aberration as the excitation beam at the BFA of an infinitely
corrected objective lens deviates from the perfectly collimated
condition40. The amount of spherical aberration introduced in this
work did not extensively affect the imaging resolution over the axial
range that we used (Supplementary Fig. 1), with the largest measured
change in lateral resolution being a factor of 1.76 and the largest change
in axial resolution being a factor of 1.52. Although such resolution
changes had a small, but noticeable, effect on structural imaging (as can
be seen in Figs. 3 and 4), the effect on functional imaging was
negligible, as the tiniest functional units we are interested in (that is,
spines) are not substantially smaller than our largest spot size. Other
groups have shown that there are minimal reductions in spherical
aberration from collimation changes in water-immersion lenses over a
similar axial range41. With larger axial ranges, however, it is generally
anticipated that spherical aberration will become more important. In
718
such cases, it might prove useful to utilize a reverse Gaussian aperture42
to preserve the resolution over the entire axial range at the expense of a
mildly reduced resolution at the inherent focal plane.
One issue that we did not address in this context is temporal pulse
broadening. Although the scanning mechanism that we employed
inherently compensates for spatial dispersion32, it does not compensate
for temporal dispersion. Consequentially, the pulse width was
broadened from 200 fs when exciting the titanium:sapphire laser to
approximately 1.8 ps at the BFA. This required an increase in laser
power for two-photon fluorescence excitation, and we used average
powers of approximately 40–100 mW at the BFA for these studies. This
power increase also compensated for the decrease in intensity at
positions that were not at the focal point of the objective because of
increased spot sizes (see Supplementary Fig. 1), which would otherwise also reduce excitation efficiency. Notably, this amount of power
was approximately 10–25% of the amount available at the BFA, thus we
were not power limited by the broadened pulse width. In addition,
although it might seem that the required higher power levels would
lead to increased photodamage, previous studies have shown equivalent levels of photodamage in multiphoton microscopy when using
pulse widths ranging from less than hundred femtoseconds to several
picoseconds43. It has even been suggested that potential photodamage
can be reduced with longer pulse widths44. For these reasons, combined
with the physical impracticality of a prism-based prechirper large
enough to compensate for the B80,000 fs2 group velocity dispersion
introduced by the four AODs, we did not use an external temporal
compensation device.
On the other hand, it has been also been suggested that using narrow
pulses in multiphoton microscopy will lead to increased penetration
depth45, as higher peak powers can be applied to deeper tissue. Thus,
with increasing axial ranges or with in vivo imaging where more tissue
is transversed46, external temporal dispersion compensation might
become increasing valuable, in which case, compensation schemes
using diffraction gratings47 or photonic crystal fibers48 offer possible
methods for generating the large amount of compensatory negative
group velocity dispersion values. Another possibility is to use moderately longer pulse lengths than those that we started with in this
experiment, which minimizes both the temporal dispersion and spatial
dispersion induced by the system. However, these optimal midrange
femtosecond pulse lengths are not widely commercially available on
most laser systems and typically require either off-label manipulations
to the laser cavity15 or custom-designed laser systems30.
Two recent studies have reported the development of other methods
for three-dimensional functional neuronal recordings, both of which
are substantially different from the method proposed here. The first
combines a scanning system that consists of galvanometer-driven
pivoting mirrors for lateral imaging and a piezoelectric actuator for
axial imaging with a new scanning strategy26. In essence, this technique
represents the furthest extent to which inertia-based scanning techniques can be taken. Indeed, when used for scanning neuronal populations, it has been shown that up to 90% of user-selected cells can be
imaged at acquisition rates on the order of 10 Hz. The remaining
fundamental limitation is the dependence on actuator- and galvanometer-based scanning technologies. In fact, even the fastest piezoelectric actuators are only capable of speeds of around 100 Hz, if not
slowed down by the inertia of a microscope objective. At the higher
speeds, however, the vibration from physical movement of the objective
lens can easily disturb image quality and disrupt biological samples. In
contrast, the AOD-based scanning approach presented here is easily
capable of acquisition rates that are in the kilohertz range, without
imparting any physical vibrations to the sample or the image. Although
VOLUME 11
[
NUMBER 6
[
JUNE 2008 NATURE NEUROSCIENCE
© 2008 Nature Publishing Group http://www.nature.com/natureneuroscience
TECHNICAL REPORT
the present 50-mm range of our scanner is smaller than the axial range
of many piezoactuators, which can be hundreds of microns, as
mentioned above, this axial range can be extended to similar values
by using lower-magnification objective lens, wide angular–aperture
deflectors or a combination of both.
The second study, which demonstrated a method for only fast threedimensional structural scanning, used two AODs to quickly scan a
volume30. This approach can be seen as a simplified version of the
scanning scheme presented here, where the acoustic frequencies
applied to the first X and Y deflectors are kept constant, or the first
two deflectors are bypassed altogether, but chirped frequencies are
applied to the latter X and Y deflectors. Because these acoustic
frequencies were chirped, the laser beam after the AODs was no longer
collimated (Fig. 1b). In addition, similar to the method presented here,
the degree of collimation, and hence the axial position, can be changed
by changing the degree of chirp. Unlike the method presented here,
however, there is no secondary set of deflectors in the path. As a result,
there is a continuous lateral movement of the laser beam that is not
compensated for and forces the scanner into a continuous ‘raster’
mode. Although this approach does allow for three-dimensional
structural imaging, it has two consequences for functional imaging.
First, it limits the effective speed of the system because time is wasted at
sites that are not of interest, which is true with all raster scanning
systems. Second, this scheme markedly reduces the dwell time per point
of interest. This severely limits the number of detected photons and
reduces the achievable signal-to-noise ratio, a fact which the authors
acknowledged. These two consequences make this approach nonoptimal for functional three-dimensional imaging. Indeed, although the
authors have shown functional imaging signals, they have been able to
do this only when not scanning in three dimensions. When acquiring
functional data, their imaging scheme resembles previously published
two-dimensional schemes15. In this regard, although both approaches
allow for fast three-dimensional structural imaging, this approach,
unlike ours, does not permit fast three-dimensional functional
imaging, where high scanning speeds are arguably more important.
In summary, we have developed a multiphoton microscope that
represents, to the best of our knowledge, the first device that is capable
of recording from multiple sites in a tissue volume at acquisition rates
that are in the tens of kilohertz range. We have used this instrument to
monitor calcium dynamics in a single neuron. Such experiments have
previously been limited to a single focal plane, which often restricts
researchers to very few sites, as dendritic processes can rapidly pass
through this plane. Our approach can also be combined with similarly
designed three-dimensional random-access photolysis of caged compounds, which opens the door for countless experiments examining the
spatiotemporal dynamics of neural computation. In addition, this
technique can be immediately applied to a variety of other neurophysiological situations, most notably in the study of cell populations. In
this case, lower magnification would be employed, and indeed, we have
previously shown an axial range of approximately 1 mm when the
AOD-based system was combined with a 10 objective, giving a large
axial range over which cells could be imaged. Given its applicability to a
wide range of neurophysiological experiments, this imaging tool
represents an important technological advancement in the current
experimental arsenal of neurophysiologists.
METHODS
Brain-slice preparation. Hippocampal brain slices were obtained from
4–6-week-old Sprague Dawley rats and prepared in accordance with the
guidelines of the US National Institutes of Health, as approved by the
Institutional Animal Care and Use Committee of Baylor College of Medicine.
NATURE NEUROSCIENCE VOLUME 11
[
NUMBER 6
[
JUNE 2008
Rats were anesthetized with a mixture of ketamine, xylazine and acepromazine
and perfused transcardially with a 2–4 1C solution consisting of 2.5 mM KCl,
1.25 mM Na2H2PO4, 25 mM NaHCO3, 0.5 mM CaCl2, 7 mM MgCl2, 7 mM
dextrose, 100 mM choline chloride, 1.3 mM ascorbate and 3 mM pyruvate. The
brain was dissected, hemisected and mounted in the holding chamber of a
tissue slicer (Vibratome 100, Ted Pella) in this same solution. Hippocampal
brain slices (350 mm thick) were cut and subsequently transferred to a holding
chamber containing an artificial cerebrospinal fluid solution containing
125 mM NaCl, 2.5 mM KCl, 1.25 mM NaH2PO4, 25 mM NaHCO3, 2 mM
CaCl2, 2 mM MgCl2 and 10 mM dextrose at 35 1C. The solution was bubbled
with a 95% O2, 5% CO2 mixture. Slices were held at 35 1C for 45 min before
moving them to a room temperature (21–24 1C) recording chamber 20 min
before use. For experiments, the solution was held at 30–34 1C using an in-line
temperature controller (TC-324B, Warner Instruments).
Electrical and optical measurements. Patch pipettes were pulled using a
pipette puller (Model P-97, Sutter Instruments) to tips with resistance values
between 3–5 MOs. These pipettes were filled with solution containing 120 mM
potassium gluconate, 20 mM KCl, 10 mM HEPES, 2 mM MgCl2, 4 mM
magnesium GTP, 0.3 mM sodium GTP and a combination of a structural label,
50 mM Alexa 594, and a Ca2+ indicator, 200 mM OGB-1 (both Molecular
Probes). Pyramidal neurons of interest from the CA1 region of the hippocampus were identified and patched in the whole-cell configuration using wide
field infrared differential-interference contrast imaging at depths approximately
50–100 mm below the surface of the slice. This was accomplished by removing
the filter slider holding the short-pass dichroic mirror (shown in Fig. 2),
allowing trans-illumination light to reach an infrared-sensitive camera
mounted on the top of the microscope. Measured changes in Ca2+ indicator
fluorescence were normalized using DF/F values, which were calculated
according to equation (4) below,
DF
F Fbaseline
¼ 100% F
Fbaseline Fbackground
ð4Þ
where F is the value of the PMT-generated voltage signal at a given time point,
Fbaseline is the average value of the prestimulus signal and Fbackground is the
average value of the measured dark noise from the PMT. Signals were low-pass
filtered to 50 Hz using a finite impulse response filter to remove residual
shot noise.
Note: Supplementary information is available on the Nature Neuroscience website.
ACKNOWLEDGMENTS
We thank Y. Liang for his contributions to the neurophysiology experiments.
This project was supported by grants from the US National Institutes of Health
and the National Science Foundation to P.S.
AUTHOR CONTRIBUTIONS
G.D.R. designed the microscope, performed the initial experiments and wrote the
manuscript. K.K. carried out the neurophysiological experiments and R.F.
designed the software. P.S. Conceived the three-dimensional scanning and
supervised the project.
Published online at http://www.nature.com/natureneuroscience
Reprints and permissions information is available online at http://npg.nature.com/
reprintsandpermissions
1. Megias, M., Emri, Z., Freund, T.F. & Gulyas, A.I. Total number and distribution of
inhibitory and excitatory synapses on hippocampal CA1 pyramidal cells. Neuroscience
102, 527–540 (2001).
2. Ambros-Ingerson, J. & Holmes, W.R. Analysis and comparison of morphological reconstructions of hippocampal field CA1 pyramidal cells. Hippocampus 15, 302–315
(2005).
3. Losonczy, A. & Magee, J.C. Integrative properties of radial oblique dendrites in
hippocampal CA1 pyramidal neurons. Neuron 50, 291–307 (2006).
4. Frick, A., Magee, J., Koester, H.J., Migliore, M. & Johnston, D. Normalization of Ca2+
signals by small oblique dendrites of CA1 pyramidal neurons. J. Neurosci. 23,
3243–3250 (2003).
5. Helmchen, F., Svoboda, K., Denk, W. & Tank, D.W. In vivo dendritic calcium dynamics in
deep-layer cortical pyramidal neurons. Nat. Neurosci. 2, 989–996 (1999).
6. Markram, H., Helm, P.J. & Sakmann, B. Dendritic calcium transients evoked by single
back-propagating action potentials in rat neocortical pyramidal neurons. J. Physiol.
(Lond.) 485, 1–20 (1995).
719
© 2008 Nature Publishing Group http://www.nature.com/natureneuroscience
TECHNICAL REPORT
7. Kovalchuk, Y., Eilers, J., Lisman, J. & Konnerth, A. NMDA receptor-mediated subthreshold Ca2+ signals in spines of hippocampal neurons. J. Neurosci. 20, 1791–1799
(2000).
8. Yuste, R. & Denk, W. Dendritic spines as basic functional units of neuronal integration.
Nature 375, 682–684 (1995).
9. Yuste, R., Majewska, A., Cash, S.S. & Denk, W. Mechanisms of calcium influx into
hippocampal spines: heterogeneity among spines, coincidence detection by NMDA
receptors and optical quantal analysis. J. Neurosci. 19, 1976–1987 (1999).
10. Sabatini, B.L. & Svoboda, K. Analysis of calcium channels in single spines using optical
fluctuation analysis. Nature 408, 589–593 (2000).
11. Sabatini, B.L., Oertner, T.G. & Svoboda, K. The life cycle of Ca2+ ions in dendritic spines.
Neuron 33, 439–452 (2002).
12. Hoogland, T.M. & Saggau, P. Facilitation of L-type Ca2+ channels in dendritic spines by
activation of b2 adrenergic receptors. J. Neurosci. 24, 8416–8427 (2004).
13. Mainen, Z.F. et al. Two-photon imaging in living brain slices. Methods 18, 231–239
(1999).
14. Salome, R. et al. Ultrafast random-access scanning in two-photon microscopy using
acousto-optic deflectors. J. Neurosci. Methods 154, 161–174 (2006).
15. Iyer, V., Hoogland, T.M. & Saggau, P. Fast functional imaging of single neurons using
random-access multiphoton (RAMP) microscopy. J. Neurophysiol. 95, 535–545
(2006).
16. Bansal, V., Patel, S. & Saggau, P. High-speed addressable confocal microscopy for
functional imaging of cellular activity. J. Biomed. Opt. 11, 34003 (2006).
17. Lv, X.H., Zhan, C., Zeng, S.Q., Chen, W.R. & Luo, Q.M. Construction of multiphoton laser
scanning microscope based on dual-axis acousto-optic deflector. Rev. Sci. Instrum. 77,
046101 (2006).
18. Lechleiter, J.D., Lin, D.T. & Sieneart, I. Multi-photon laser scanning microscopy using an
acoustic optical deflector. Biophys. J. 83, 2292–2299 (2002).
19. Saggau, P. New methods and uses for fast optical scanning. Curr. Opin. Neurobiol. 16,
543–550 (2006).
20. Zhu, L., Sun, P.C. & Fainman, Y. Aberration-free dynamic focusing with a multichannel
micromachined membrane deformable mirror. Appl. Opt. 38, 5350–5354
(1999).
21. Oku, H., Hashimoto, K. & Ishikawa, M. Variable-focus lens with 1-kHz bandwidth. Opt.
Express 12, 2138–2149 (2004).
22. Oron, D., Tal, E. & Silberberg, Y. Scanningless depth-resolved microscopy. Opt. Express
13, 1468–1476 (2005).
23. Durst, M.E., Zhu, G.H. & Xu, C. Simultaneous spatial and temporal focusing for axial
scanning. Opt. Express 14, 12243–12254 (2006).
24. Babin, A.A., Kartashov, D.V. & Kulagin, D.I. Focusing femtosecond radiation with an
axicon. IEEE J. Quantum Electron. 32, 308–310 (2002).
25. Dufour, P., Piche, M., De Koninck, Y. & McCarthy, N. Two-photon excitation fluorescence
microscopy with a high depth of field using an axicon. Appl. Opt. 45, 9246–9252
(2006).
26. Gobel, W., Kampa, B.M. & Helmchen, F. Imaging cellular network dynamics in three
dimensions using fast 3D laser scanning. Nat. Methods 4, 73–79 (2007).
27. Reddy, G.D. & Saggau, P. Fast three-dimensional laser scanning scheme using acoustooptic deflectors. J. Biomed. Opt. 10, 064038 (2005).
720
28. Iyer, V., Losavio, B.E. & Saggau, P. Compensation of spatial and temporal dispersion for
acousto-optic multiphoton laser-scanning microscopy. J. Biomed. Opt. 8, 460–471
(2003).
29. Zeng, S. et al. Simultaneous compensation for spatial and temporal dispersion of
acousto-optical deflectors for two-dimensional scanning with a single prism. Opt. Lett.
31, 1091–1093 (2006).
30. Vucinic, D. & Sejnowski, T.J. A compact multiphoton 3D imaging system for recording
fast neuronal activity. PLoS. ONE 2, e699 (2007).
31. Bullen, A., Patel, S.S. & Saggau, P. High-speed, random-access fluorescence microscopy. I. High-resolution optical recording with voltage-sensitive dyes and ion indicators.
Biophys. J. 73, 477–491 (1997).
32. Bullen, A. & Saggau, P. High-speed, random-access fluorescence microscopy. II. Fast
quantitative measurements with voltage-sensitive dyes. Biophys. J. 76, 2272–2287
(1999).
33. Shoham, S., O’Connor, D.H., Sarkisov, D.V. & Wang, S.S.H. Rapid neurotransmitter
uncaging in spatially defined patterns. Nat. Methods 2, 837–843 (2005).
34. Xu, J. & Stroud, R. Acousto-optic Devices: Principles, Design and Applications (John
Wiley and Sons, New York, 1992).
35. Gottlieb, M., Ireland, C.R.E. & Ley, J.M. Electro-Optic and Acousto-Optic Scanning and
Deflection (Marcel Dekker, New York, 1983).
36. Kaplan, A., Friedman, N. & Davidson, N. Acousto-optic lens with very fast focus
scanning. Opt. Lett. 26, 1078–1080 (2001).
37. Siegman, A.E. Lasers (University Science Books, Sausalito, California, 1986).
38. Xu, J., Weverka, R.T. & Wagner, K.H. Wide angular–aperture acousto-optic devices.
Proc. Soc. Photo. Opt. Instrum. Eng. 2754, 104–114 (1996).
39. Chang, I.C. Acousto-optic modulator with wide bandwidth and large angular aperture.
Electron. Lett. 30, 1190–1191 (1994).
40. Sheppard, C.J.R. & Gu, M. Aberration compensation in confocal microscopy. Appl. Opt.
30, 3563–3568 (1991).
41. Kam, Z., Agard, D.A. & Sedat, J.W. Three-dimensional microscopy in thick biological
samples: a fresh approach for adjusting focus and correcting spherical aberration.
Bioimaging 5, 40–49 (1997).
42. Escobar, I., Saavedra, G., Martinez-Corral, M. & Lancis, J. Reduction of the spherical
aberration effect in high numerical-aperture optical scanning instruments. J. Opt. Soc.
Am. A Opt. Image Sci. Vis. 23, 3150–3155 (2006).
43. Koester, H.J., Baur, D., Uhl, R. & Hell, S.W. Ca2+ fluorescence imaging with pico- and
femtosecond two-photon excitation: signal and photodamage. Biophys. J. 77,
2226–2236 (1999).
44. Hopt, A. & Neher, E. Highly nonlinear photodamage in two-photon fluorescence
microscopy. Biophys. J. 80, 2029–2036 (2001).
45. McConnell, G. Improving the penetration depth in multiphoton excitation laser scanning
microscopy. J. Biomed. Opt. 11, 54020 (2006).
46. Helmchen, F. & Denk, W. Deep tissue two-photon microscopy. Nat. Methods 2,
932–940 (2005).
47. Treacy, E.B. Optical pulse compression with diffraction gratings. IEEE J. Quantum
Electron. 5, 454–458 (1969).
48. Mogilevtsev, D., Birks, T.A. & Russell, P.S. Group-velocity dispersion in photonic crystal
fibers. Opt. Lett. 23, 1662–1664 (1998).
VOLUME 11
[
NUMBER 6
[
JUNE 2008 NATURE NEUROSCIENCE