CHAPTER

Correlative In Vivo 2-Photon

Imaging and Focused Ion
Beam Scanning Electron
Microscopy: 3D Analysis of
16
Neuronal Ultrastructure
Bohumil Maco*,{, Anthony Holtmaat{, Anne Jorstad{, Pascal Fua{
Graham W. Knott*
*BioEM Facility, Centre of Electron Microscopy, EPFL, Lausanne, Switzerland
{
Computer Vision Laboratory, EPFL, Lausanne, Switzerland
{
Department of Neuroscience, University Medical Center (CMU), University of Geneva,
Geneva, Switzerland

CHAPTER OUTLINE
Introduction ............................................................................................................ 340
16.1 Method..........................................................................................................342
16.1.1 Fixation and Vibratome Sectioning .............................................. 342
16.1.2 Laser Fiducial Marks Around Region of Interest
(NIRB Technique) ..................................................................... 344
16.1.2.1 Method ............................................................................ 344
16.1.3 Staining and Resin Embedding................................................... 345
16.1.3.1 Method ............................................................................ 345
16.1.4 Preparation of the Block Face and Mounting the Block
for FIBSEM............................................................................... 346
16.1.4.1 Method ............................................................................ 346
16.1.5 FIBSEM Imaging ....................................................................... 349
16.1.5.1 Method ............................................................................ 350
16.1.6 3D Reconstruction and Analysis ................................................. 350
16.1.6.1 Method ............................................................................ 353
Conclusion ............................................................................................................. 359
Acknowledgments ................................................................................................... 359
References ............................................................................................................. 360

Methods in Cell Biology, Volume 124, ISSN 0091-679X, http://dx.doi.org/10.1016/B978-0-12-801075-4.00016-1

© 2014 Elsevier Inc. All rights reserved.
339
340 CHAPTER 16 2-Photon Imaging and FIBSEM

Abstract
This protocol describes how dendrites and axons, imaged in vivo, can subsequently be analyzed
in 3D using focused ion beam scanning electron microscopy (FIBSEM). The fluorescent struc-
tures are identified after chemical fixation and their position highlighted using the 2-photon laser
to burn fiducial marks around the region. Once the section has been stained and resin embedded,
a small block is trimmed close to these marks. Serially aligned EM images are acquired through
this region, using FIBSEM, and the neurites of interest then reconstructed semi-automatically
using the Ilastik software (ilastik.org). This fast and reliable imaging and reconstruction tech-
nique avoids the use of specific labels to identify the features of interest in the electron micro-
scope and optimizes their preservation for high-quality imaging and 3D analysis.

INTRODUCTION
Visualizing the morphology and connectivity of neurons is crucial to understanding
how they function. The increasing use of live-imaging methods has enabled neuro-
scientists to study neurons in their native environment. However, this has also
demanded greater use of electron microscopy to give further interpretation to the
changes seen with the light microscope. With its higher resolution and ability to give
more structural details, electron microscopy remains the only technique that is able to
see fine cellular architecture and synaptic connectivity of neurons. In recent years,
the development of new EM methods has enabled large volumes of tissue to be im-
aged with only limited human interaction. These have opened up many different op-
portunities for exploring the brain.
This chapter gives a detailed explanation of one technique that has recently been
used in a number of high-profile studies combining the imaging of neurons in the live
brain, using 2-photon microscopy, with electron microscopy using the focused ion
beam scanning electron microscope (Allegra Mascaro et al., 2013; Cane, Maco,
Knott, & Holtmaat, 2014; Canty et al., 2013; Grillo et al., 2013; Mostany et al.,
2013). These studies carried out extensive series of live-imaging experiments show-
ing how the morphology of axons and dendrites of fluorescently labeled neurons
changed during aging and in response to injury.
Previous correlative light and electron microscopy (CLEM) experiments on in
vivo-imaged neurons had used serial-section transmission electron microscopy
(Holtmaat, Wilbrecht, Knott, Welker, & Svoboda, 2006; Knott, Holtmaat,
Trachtenberg, Svoboda, & Welker, 2009; Knott, Holtmaat, Wilbrecht, Welker, &
Svoboda, 2006; Trachtenberg et al., 2002; Zito, Parnas, Fetter, Isacoff, &
Goodman, 1999), however, these studies were rare, requiring considerable amounts
of time and skill to manually section large volumes of tissue that is then imaged one
section at a time to locate the region of interest. Maintaining a complete series of
sections, in perfect condition, is critical for this method to work. The first study
of this kind appeared soon after green fluorescent protein (GFP) started to be used
in live-cell imaging, and was undertaken in Drosophila (Zito et al., 1999) to under-
stand how the neuromuscular junction developed and synapses inserted into the
 Introduction 341

growing boutons. This pioneering work required the axons of interest to be found in
the large series of sections without any label being used to highlight its whereabouts.
Later studies of neuronal plasticity in the mammalian adult brain used antibodies
against the GFP in a preembedding protocol that revealed the location of the structure
of interest in the resin-embedded tissue. This also allows specific neurites to be found
in the bewilderingly complex networks of the central nervous system.
Labeling fluorescent cells, so that they can be found later with electron micros-
copy, however, risks being counterproductive as the label itself may obscure the ul-
trastructural details. For this reason, the method outlined here has been developed
using the near-infrared branding (NIRB) method (Bishop et al., 2011; Maco et al.,
2013) to make fiducial marks inside the tissue close to the structures of interest,
but not associated with them. Using a 2-photon laser, the fixed tissue, with the fluo-
rescent structures still visible, is branded with easily identifiable marks. These are
seen both in the resin-embedded blocks and in the electron microscope and are used
to locate the position of the features of interest. This region is then imaged with FIB-
SEM, making sure that the entire volume indicated by the NIRB marks is captured.
This approach has several advantages to previous correlative methods. Without
using any immunolabeling converting the fluorescence into an electron-dense signal
ensures that optimal fixation protocols can be used without compromising the quality
of the ultrastructure. Using FIBSEM provides an automated imaging method requir-
ing only limited human intervention to obtain large series of uninterrupted images.
This block-face scanning approach not only gives aligned images, but as the ion
beam is able to remove a few nanometers from the block surface, image stacks with
isotropic voxels of only a few nanometers in each dimension are possible (see, e.g.,
Fig. 16.5). As well as giving a more accurate view of 3D structures, it also provides
better opportunities to use computer vision algorithms for automatically segmenting
and reconstruction of certain features (Kreshuk et al., 2011; Straehle, Köthe,
Knott, & Hamprecht, 2011).
With the continued development of block-face scanning electron microscopy, de-
tailed analysis using manual methods on ever-larger series of images is becoming
unfeasible. The advances in computer vision algorithms have made significant
strides toward automated methods for recognizing and reconstructing different fea-
tures, particularly in serial FIBSEM images (Becker, Ali, Knott, & Fua, 2013;
Lucchi, Smith, Achanta, Knott, & Fua, 2012; Lucchi, Smith, Achanta, Lepetit, &
Fua, 2010). This includes structures such as mitochondria, synapses, and neurites.
This chapter describes how, once the correct region has been imaged, axons and den-
drites can be reconstructed automatically from the serial images using the Ilastik
software (www.ilastik.org). This task would take a few days of work if attempted
by hand. Instead, it is now possible in a few hours to generate a computer model
of the structures of interest. This is then ready for further analysis. The final part
of this chapter describes how measurements can be made of these models using a
set of software tools that run on the open-source 3D imaging platform Blender
(www.blender.org). Measuring the sizes of structures like synaptic contacts, den-
dritic spines, and axonal boutons is an important means of comparing different
342 CHAPTER 16 2-Photon Imaging and FIBSEM

neurites, to understand how they function. These tools offer an accurate and rapid
means to make morphometric measurements in a 3D workspace, with rapid access
to the serial images from which the models were derived.
This complete protocol represents a significant advance in the field of correlative
light and electron microscopy. As well as using optimal fixation and staining con-
ditions for this type of material, the imaging is automated, as well as the reconstruc-
tion process. This is a method that is therefore not only more reliable but faster and
more accurate.

16.1 METHOD
16.1.1 FIXATION AND VIBRATOME SECTIONING
The fixation of the living brain is the most crucial part of the entire procedure to en-
sure high-quality EM images. The fixative is perfused through the circulatory system
as soon after opening the chest cavity as possible, with the animal deeply anesthe-
tized throughout. The brain is therefore exposed to fixatives without it experiencing
hypoxic conditions for more than 30 s. Any delay in this part will result in alterations
to the ultrastructure (Tao-Cheng, Gallant, Brightman, Dosemeci, & Reese, 2007).
The fixation conditions are considered very standard and have changed little from
the original EM fixation studies in the 1960s (Schultz & Karlsson, 1965). The pro-
cess is complete after 2 h, and therefore, the brain can be removed and vibratome
sectioned shortly after fixation.
Vibratome sections need to be cut parallel to the imaging plane of the 2-photon
microscope to ensure that the orientation of the neurites in the fixed material is sim-
ilar to those seen in the live brain. If the orientation is too dissimilar, recognizing the
same neurites in the final EM image stack is too difficult.
The protocol begins once the live 2-photon imaging has been completed with the
animal deeply anesthetized, nonreflexive, and ready to be perfusion fixed.
Animal perfusion fixation
1. At the end of the 2-photon in vivo imaging session, take a low-magnification
image of the blood vessel vasculature through the cranial window (Fig. 16.1A) to
ensure that the cell of interest can be located relative the blood vessel pattern.
2. Pin the mouse to the dissection board, exposing its underside, and then, without
delay, cut into the chest cavity to reveal the heart and lungs using surgical scissors.
3. Push the blunt perfusion needle (100 Straight Cannula Blunt End Tip,
15 gauge; Fisnar) into the left ventricle, taking care not to damage the walls of
the heart. Fix the needle in place with a heart clamp.
4. Perfuse the animal with 10 ml of isotonic PBS, with a perfusion pump, at
rate of 12 ml/min followed immediately with fixative solution. To avoid
stopping and restarting the pump to change solutions, backfill the tubing
with PBS, after the entire tubing is initially filled with fixative.
5. Perfuse the animal with 300 ml of 2.5% glutaraldehyde and 2%
paraformaldehyde in 0.1 M phosphate buffer (pH 7.4). The perfusion takes
 16.1 Method 343

A B

C D

FIGURE 16.1
In vivo imaging and near-infrared branding (NIRB) of the fiducial laser marks. (A) Blood
vasculature on the surface of the brain, seen through the cranial window. The pattern of
the blood vessels (white dotted lines) is seen in 2-photon microscope (B) and is used to locate
the region of interest (white square). (C) High-magnification image of the region of
interest with GFP-labeled neurites indicated by white square in (A) and (B). (D) 2-Photon
laser was used to burn the fiducial laser marks (white arrowheads) around the structure of
interest on the fixed vibratome section containing the region of interest previously imaged
in vivo. Scale bar in (A) and (B) is 100 mm, and in (C) and (D) is 10 mm.
Modified from Maco et al. (2013); reproduced with permission.

approximately 25 min, and once completed, the animal is then left for a further
90 min at room temperature, and sealed inside a plastic bag.
6. Dissect the brain from the skull using thin, blunt forceps, taking care not to
damage the brain tissue. Once the brain is removed immerse in PBS.

Vibratome sectioning
7. Sink the brain in molten agarose (5%, in PBS) that has been cooled to 30  C, and
manipulate its position until the region of the cranial window is approximately
344 CHAPTER 16 2-Photon Imaging and FIBSEM

positioned uppermost. Then place in the fridge for 10 min or until the
agarose sets.
8. Trim the agarose block, containing the brain, so that the area of cortex imaged
through the cranial window is parallel to the upper surface.
9. Glue the agarose block to the vibratome support using cyanoacrylate glue.
10. Vibratome section 60-mm thick slices (Leica VT1200S, Leica Microsystems)
from the region of the cranial window making sure that the sections are cut in
the same plane as the focal plane as the 2-photon microscope. Cut sections
picked up with a paint brush and place sequentially in a 24-well culture plate.

16.1.2 LASER FIDUCIAL MARKS AROUND REGION OF INTEREST

(NIRB TECHNIQUE)
As the protocol relies on making fiducial marks in the fixed tissue, it is crucial that
there is enough visible fluorescent signal after using this type of fixative with a high
concentration of glutaraldehyde (2.5%). Without any label to see the presence of the
fluorescent signal in the electron microscope, the neurites of interest need to be pin-
pointed together with fiducial marks. This gives their relative position allowing them
to be found in the serial images.
The number of NIRB marks made depends on how many features are already pre-
sent in the slice. If there is a clear pattern of blood vessels, for example, it would only be
necessary to make a single square around the neurites. However, if natural landmarks
are less obvious, it may be necessary to include larger burn marks around the first. These
help only in finding the small set of marks in the resin-embedding section. It may also be
useful to make NIRBs that form an asymmetric shape which will indicate from which
side the section should be imaged and attached to the block resin slab.

16.1.2.1 Method
Localization of neurites of interest on the fixed vibratome section
1. Using a dissecting microscope (Leica M205C), capture images at approximately
12  zoom to visualize the blood vessels. These are most prominent in
the first sections cut from the surface of the brain.
2. Overlay each image, into a stack, and align them (e.g., Photoshop, ImageJ, or
GIMP). This stack will reveal the complete pattern of blood vessels on the brain’s
surface.
3. Match the image of the blood vessels from the fixed sections with the image of
the blood vessel pattern taken of the live brain through the cranial window.
This will reveal in which section to find the fluorescent structures of interest.
4. Place the identified section on a glass microscope slide, with adequate PBS to
ensure that it does not dry. Place a coverslip on top and seal with nail varnish.
Making the NIRB marks
5. Identify the neurites of interest (Fig. 16.1B and C) using the 2-photon
microscope. Then set the laser to a wavelength of 850 nm, with a power of
 16.1 Method 345

300 mW at the back focal plane of the objective. Here, a 2-photon laser scanning
microscope equipped with a pulsed titanium sapphire laser (Chameleon Ultra II,
Coherent) was used.
6. In the line scan mode, with a speed of approximately 2 ms/line, burn stripes in the
fixed tissue around the region of interest. Between 500 and 2000 sweeps were
used until the marks became visible from the autofluorescence generated from
the line edges (Fig. 16.1D). The more sweeps used, the wider the burnt stripe, and
the further apart the two fluorescent edges. Make sure that the photomultipliers
are switched off for this marking phase.
7. NIRB marks are made around the neurites of interest in the form of a square or
rectangle approximately 15–24 mm along one side (Fig. 16.1D). These are also
placed approximately 3 mm above the neurites of interest in the Z-axis.
8. Image this region, with the NIRB marks, using the 2-photon laser, and then again
using transmitted light, to capture its position in relation to other features, such as
blood vessels and cell bodies.

16.1.3 STAINING AND RESIN EMBEDDING

The staining and resin embedding part of this protocol uses a standard method with
conventional reagents. This approach has been used for years by many laboratories.
The most important consideration is to stain the membranes as fully as possible to
give good contrast with FIBSEM, while being careful not to overstain the tissue that
would make it difficult to view the laser marks in the embedded section using trans-
mitted light microscopy. Therefore, the timings are important. Although thinner sec-
tions could be considered as a solution, these often do not remain flat during the
dehydration process and end up with an undulating appearance in the final hardened
resin. This gives regions of interest that are difficult to mount horizontally on the
blank resin slab.

16.1.3.1 Method
Staining, dehydration, and resin embedding of the fixed sections
1. Wash the section 3  for 5 min each change in 0.1 M sodium cacodylate buffer
(pH 7.4), in a 20 ml glass scintillation vial mixing continuously on a rotating
mixer.
2. Stain in 1% osmium tetroxide and 1.5% ferrocyanide in 0.1 M sodium
cacodylate buffer (pH 7.4) for 40 min.
3. Stain with 1% osmium tetroxide in 0.1 M sodium cacodylate buffer (pH 7.4) for
40 min.
4. Rinse the sections in ddH2O, 2 for 5 min each rinse.
5. Stain section with 1% uranyl acetate in ddH2O for 40 min.
6. Rinse section in ddH2O, 5 min.
7. Dehydrate in increasing concentrations of ethanol: 2  in 50%, 1  in 70%,
1 in 90%, 1 in 95%, and 2  in 100% ethanol, 3 min each change.
346 CHAPTER 16 2-Photon Imaging and FIBSEM

8. Infiltrate the sections with 1:1 mixture of Durcupan resin and 100% ethanol for
30 min.
9. Infiltrate the section for 2  30 min with fresh resin each change.
10. Exchange for fresh resin and agitate for 4 h.
11. Arrange the section onto treated glass slide (coated with mold separating agent),
add more resin and overlay a second glass slide to sandwich the section
between.
12. Polymerize the resin and section in 65  C oven for 24 h and then separate the
glass slides from the resin using a razor blade.

16.1.4 PREPARATION OF THE BLOCK FACE AND MOUNTING

THE BLOCK FOR FIBSEM
The goal of the block preparation process is to have a block face in the microscope on
which the observer can pinpoint the exact position of the NIRB marks and without
having to undertake long periods of milling before the final image series is collected.
This requires trimming with glass knives mounted in the ultramicrotome so that the
NIRB marks lie close to the edge of the resin block and within 4 mm of its surface
(Fig. 16.3F).
To help guide the trimming process, a laser-dissecting microscope can also be
used to etch the block surface and mark the position of the NIRB marks within
the block (Fig. 16.2C). These surface marks are visible with the stereomicroscope
and also in the ultramicrotome. This means that the block does not need to be re-
moved from its mounting and imaged under a transmitted light microscope to check
each time the position of the required region.
To locate the exact region to be imaged with FIBSEM, a small indentation is also
made each side of the final imaging face (Fig. 16.3C), as well as score marks running
across the top of the block (Fig. 16.3D). The position of these marks are also carefully
measured, in relation to the NIRB marks, as they will be the only indicators with
which to place the final imaging window for the collection of the FIBSEM image
stack.

16.1.4.1 Method
1. Image the resin-embedded section in the light microscope to pinpoint the
position of the NIRB marks (Fig. 16.2B).
2. Etch the shape of the NIRB onto the surface of the resin section, using the pulsed
laser of the laser dissection microscope (Fig. 16.2C). These marks serve to
indicate the position of the NIRB marks inside the resin block.
3. Cut a 4  20 mm strip of blank resin, with thickness of 1 mm, using a jeweler’s
saw, and rinse with water and then dry, to remove all the dust particles.
4. Glue the section of interest, containing the NIRB mark, to the resin slab with
a small drop of Durcupan resin and polymerize in the oven. The drop of
resin should be small so that it does not raise the section too much from the
surface of the slab (Figs. 16.2D and 16.3A).
 16.1 Method 347

A B

C D

FIGURE 16.2
Tissue preparation for the FIBSEM. (A) Fiducial laser marks can be seen inside the fixed
tissue section (black arrowheads) as well as after staining and embedded in resin (B, white
arrowheads). (C) Using a laser-dissecting microscope, the original NIRB marks were
replicated on to the top surface of resin-embedded sample to help indicate the region of
interest during the trimming process. (D) The piece of resin, with the NIRB marked region of
interest, was cut from the section and mounted onto a flat slab of empty resin providing a
thicker support for easy manipulation during the trimming process. Note the asymmetric
shape of the 2-photon laser marks used to distinguish the correct orientation, as well as the
larger laser marks, used to facilitate the search and identification of the region of interest
on the resin-embedded section. Scale bar: 20 mm.

5. Clamp the block in the ultramicrotome so that the embedded section with NIRB
mark is held vertically.
6. Trim away the resin, using a glass knife, so that the region of interest marked
by the NIRB marks lies less than 10 mm from this front face (Fig. 16.3B).
Use the etched surface of the slab to guide the trimming process, and avoid too
much resin being removed.
7. Rotate the block so that the NIRB region is lying horizontally (Fig. 16.3C)
and trim an indentation of <50 mm to the right and left of the region of interest,
348 CHAPTER 16 2-Photon Imaging and FIBSEM

A B

C D

E F Ion beam direction

e– beam
direction
*

FIGURE 16.3
Trimming process and preparation of the block face for the FIBSEM. (A) Piece of resin-
embedded section with 2-photon laser marks (black arrowhead) replicated on the top surface
(white arrowheads) by the laser dissection microscope and mounted on the slab of empty
resin. (B) The slab is then mounted on the ultramicrotome, and using the glass knife, one
side is trimmed away until the NIRB marks are exposed at the edge. (C) To locate the region of
interest inside the FIBSEM, indentations are made either side of the laser marks and (D)
two score marks are made down the front edge (black arrowheads) serving as additional
fiducial marks. These are easily seen in the electron microscope. (E) The block is then
orientated to allow any excess resin above the NIRB marks to be removed. (F) The block
is then mounted on a 45 inclined aluminum SEM stub, and coated with a protective layer
of gold (approximately 30 nm thick). Inside the FIBSEM, the focused ion beam is
oriented parallel to the block face (black arrow) and about 54 relative to the electron beam
(white arrow).
 16.1 Method 349

about 50 mm from the NIRB marks (Fig. 16.3C). Additionally, score marks
can be made down the face of the block marking the position of the NIRB
mark in the block face. This is achieved using the edge of the glass knife
held at an angle of 45 (Fig. 16.3D).
8. Hold the resin block in the jaws of the ultramicrotome holder so that the surface
of the block, parallel to the NIRB mark, is also parallel to the edge of the
glass knife. Trim away the resin from above the NIRB marks (Fig. 16.3E).
Ensure that only approximately 5 mm of resin remains above the NIRB marks.
Mounting the block to SEM stub and gold coating
9. Photograph this final block, using the light microscope, to show the position
of the NIRB mark inside the resin block relative to the block’s edges, the
indentation made on the right-hand side of the block, and the score marks
on the front edge.
10. Match this final block image with the 2-photon image of the NIRB marks
that also includes the structures of interest. These two images are used to
precisely position the imaging face in the FIBSEM.
11. Glue the final trimmed block to the scanning electron microscopy (SEM)
stub, on the included surface, using carbon conductive cement.
12. Place the stub in the sputter coater and deposit a 30 nm layer of gold over
the surface.

16.1.5 FIBSEM IMAGING

FIBSEM images are taken from the resin-embedded tissue, at the same angle as the
2-photon images were collected in the live brain. This means that the block face is
milled with the ion beam at the same plane as the focus plane of the 2-photon mi-
croscope (Fig. 16.3F). This results in imaging of the neurites of interest at the same
orientation as they were imaged live.
The initial work in the FIB microscope is to roughly mill the block face according
to the position indicated by the score marks, described above. Once the underlying tis-
sue is exposed milling continues until other fiducial marks are seen, such as the NIRB
marks. Once this has been achieved the block face can then be finally polished and a
protective layer is deposited on the top surface of the block. This layer of platinum or
carbon ensures a consistent milling without streaking and the curtaining effect.
Unfortunately, at this point it is not advised to immediately start capturing the
final image series. During these milling steps, to prepare the final imaging window,
the block has been heated and will drift while a thermal equilibrium is reached. For
this reason, it is advisable to switch off the electron beam, close the valve of the ion
beam, and leave the microscope for at least 2 h. During this time, a thermal and me-
chanical stability will have been reached and less drift of the block will occur.
During the imaging phase, accurate focus and consistent milling thickness is par-
amount. Although the thickness is given by the initial settings of the ion beam, in
reality this can be different and needs to be calculated post hoc, using the cylindrical
diameters method (Fiala & Harris, 2001).
350 CHAPTER 16 2-Photon Imaging and FIBSEM

16.1.5.1 Method
1. Load the sample into the FIBSEM microscope.
2. Position the ion beam so that it is parallel to the block face. As the face is already
inclined at 45 , the SEM stub needs only to be tilted by around 10 , using
the microscope stage to ensure the ion beam is parallel to this face. This imaging
face is scanned with the electron beam during image acquisition.
3. Locate the position of the NIRB marks along the imaging face using the images
acquired during the block preparation.
4. Mill this region of the face, about 60 mm wide, with an ion beam current of
approximately 6 nA. The region of interest should be positioned in the center.
Continue milling into the block until some underlaying features in the tissue
appear when imaging the face in the back-scattered imaging mode. These
features, like blood vessels, or cells bodies, or even the NIRB marks themselves
are used to calculate the exact position of serial milling and image acquisition.
5. Deposit a protective layer of platinum (or carbon, 1-mm thick) onto the top
surface of the block, above the region of interest. This patch of deposited
material is approximately 30 mm wide and 10 mm deep (Fig. 16.4C).
6. Finely mill the imaging face with a weaker ion beam of current 0.7–1.5 nA. This
creates a more vertical and smooth face.
7. Choose the exact region for sequential milling and imaging (approximately
15  10 mm) to capture the neurites of interest (Fig. 16.4).
8. Set the ion beam current to 700 pA with a voltage of 30 kV and set the width of
the milling window to three times wider than the imaged face. This ensures that
no redeposition occurs on the face to be imaged.
9. With an electron beam current of 800 pA, voltage of 1.5 kV, and grid tension on
the ESB detector (Zeiss microscopes only) of 1.3 kV start imaging the block face.
10. Acquire images with pixel size of 4–5 nm per pixel and dwell times of
around 10 ms. Mill the face with a thickness of around 5–10 nm each slice.
11. Use microscope dynamic focus, image tracking, and tilt correct functions to
keep the images focused while the block face retreats during sequential milling,
and in focus down the entire face of the tilted sample.

16.1.6 3D RECONSTRUCTION AND ANALYSIS

The analysis part begins with finding the neurites in the serial images. This can be
achieved simply by overlaying the first EM image onto the fluorescent NIRBs image
that also includes the neurites of interest. This will show the location, in x and y, at
which the different neurites will be seen in the EM image stack.
The 3D models of the neurites of interest are generated from the image stack with
the Ilastik software (Ilastik.org). This uses advanced watershed algorithms and al-
lows the user to carry out segmentations of selected features, and importantly, to rap-
idly make corrections to the generated 3D models when mistakes are made (Ilastik.
org; Fig. 16.5A and B). The small spacing in the Z direction optimizes the accuracy
of the reconstruction and reduces the number of errors. Here, the protocol explains
 A

FIGURE 16.4
FIBSEM imaging. (A) A model representing the orientation of the block inside the FIBSEM
(left) and resin-embedded tissue (right) with visible NIRB laser marks (white arrowheads).
(B) The block face showing the edge of the block and the trench milled out by the ion
beam imaged by back-scattered electrons showing the embedded tissue in the block. The top
face of the block is coated with a layer of carbon (arrowheads). (C) The milled face covers
the entire region of the NIRB marks (arrowheads) and the images were collected from a
smaller area contained within (indicated with the dotted square). Scale bars: in (A) is 20 mm,
and in (B) and (C) is 10 mm.
352 CHAPTER 16 2-Photon Imaging and FIBSEM

A B

C D

FIGURE 16.5
Segmentation and 3D reconstruction. FIBSEM imaging results in aligned serial images
(A) producing the image volume with near-isotropic voxel properties (B) through which the
 16.1 Method 353

how seeds are scribbled into the neurites of interest on a single image from any or-
thogonal plane, with background seeds added to indicate the regions for exclusion.
The program grows these seeds until they fill the entire volume of the selected neurite
and a reconstruction is then produced within a few seconds. This can be refined and
corrected by adding additional seeds.
The final 3D model is exported in a wavefront .obj file format that can be
imported into the Blender modeling software (www.blender.org). A set of tools have
recently been constructed for this software, named NeuroMorph, that enable the
measurement and analysis on the models in the 3D space (http://cvlab.epfl.ch/
NeuroMorph; http://wiki.blender.org/index.php/Extensions:2.6/Py/Scripts/Neuro_
tool). The toolset is composed of three tools or add-ons: Import, Measurement tools,
and Image Stack Interactions. The Import tool opens the models in the Blender soft-
ware with the correct size, calibrated to known units (microns). The Measurement
tool allows the models to be analyzed and such parameters as surface area, volume,
and length to be made of any part. The measurements are stored together with each
object, and when exported appear in a single-character separated values (CSV) text
file (.txt), hierarchically organized. Each value appears as a child of the original ob-
ject that can be easy ordered in a spreadsheet program.
Finally, the Image Stack Interactions tool allows the models to be viewed in 3D,
but also with the original EM images. These are placed at the correct position relative
to the model. The user can scroll through the images and all images are shown at their
current position in the z depth of the volume. Additionally, the image plane can also
be used to select other objects that may have been reconstructed, but are not visible.
This allows the user to search for other objects, using the EM images are maps.

16.1.6.1 Method
Alignment of the serial images
1. Using open-source free FIJI software package (http://fiji.sc/wiki/index.php/
Fiji), combine individual sequential images in tiff image format into a single
stack by selecting “File”—“Import”—“Image Sequence” command.
2. Select image slice in the middle of the stack as reference for stack alignment
by selecting “Plugins”—“Registration”—“StackReg” command using the
transformation option “translation only.”

FIGURE 16.5 —Cont’d features of interest can be visualized using arbitrary imaging planes.
(C) The Ilastik program was used to interactively segment the structures of interest using
seeded watershed segmentation. An image from the stack shows green marks made by the
user to indicate the structures of interest (object) and red marks for the background. The
yellow rectangle indicates the part of the segmented structure of GFP-labeled dendrite shown
on 2-photon image (D). (E) After the segmentation process, the 3D models of the
reconstructed dendrite (green) and axon (blue) are displayed inside the Ilastik program.
(F) The final model of the GFP-labeled neurites is rendered using the Blender 3D modeling
software (in gray is the dendrite; axons are in green and orange). Scale bar in (D) is 5 mm.
Modified from Maco et al. (2013); reproduced with permission
354 CHAPTER 16 2-Photon Imaging and FIBSEM

3. If necessary, crop the aligned stack by using Rectangular selection tool and
“Image”—“Crop” command.
4. Invert image contrast by selecting “Edit”—“Invert” command and adjust image
contrast by selecting “Image”—“Adjust”—“Brightness/Contrast” command.
5. Save processed stack as sequence of individual 8 bit, gray-scale images by
selecting “File”—“Save As”—“Image Sequence” command.
6. Determine the exact section thickness to measure the cylindrical object (e.g.,
mitochondria) diameter by dividing the number of slices it spans using the
cylindrical diameters method (Fiala & Harris, 2001).
7. Identify the exact position of the neurites of interest within the image volume by
overlaying and aligning their 2-photon image including the NIRB marks with
the known position of the NIRB marks on the block face.
Semi-automated segmentation with Ilastik
8. Segment the structure of interest and create a 3D model using the seeded
watershed segmentation algorithm within the Ilastik software (version 0.5.12;
http://ilastik.org). Create the new project by selecting “Project” tab—“New.” In
“Project” window, select “Load Stack” and choose the folder containing the
sequential images of the aligned stack. The entire stack is displayed in
orthogonal view with the possibility to scroll through the stack in any of the x–y,
x–z, or y–z planes.
9. To specify a set of features required to identify the membrane boundaries select
the “Classification” tab—“Select Features” button. In the opened “Spatial
Features” window, select “Texture” (includes the Eigenvalues of the Hessian
matrix as a membrane boundary indicator). Set the scale of the features to
“Medium” (representing the sigma value of 1.0). The selected set of features for
the entire image stack is calculated.
10. Switch to “Seeded Watershed” tab and import previously calculated
Eigenvalues of Hessian matrix of Gaussian Sigma Channel 2 as chosen “Input
weights.” In the “Select Border Indicator” window, specify whether the
boundaries in “Calculated Weights” are dark or bright. The supervoxels graph
used for watershed segmentation for the entire image stack is calculated.
11. Select “Seed 1” (as default in green color) in “Seeds” preview window, adjust
brush size (“M” key increase and “N” key decrease) and draw a line inside
the structure of interest. Select the “Background” (as default in red color) and
mark the outside of the selected structure or background seed (Fig. 16.5C).
Note that the lines representing the seeds can be placed from any displayed
planes of the orthogonal view. Additional seeds (in different colors) can be
added using the “Create Seed” button or deleted by right-clicking on the
selected seed in the list on the “Seeds” preview window.
12. Select the “Segment” button to start the segmentation. The selected structure
is colorized as “Segmentation” overlay in the “Overlays” preview window.
The segmentation can be iteratively refined by placing more foreground and
background seeds, or adding seeds for other objects. The “Bias” dialogue field
 16.1 Method 355

allows to tune the growth of the background seeds more (values around 0.8) or
less (values around 0.97) aggressively than the default.
13. Display the rendered 3D model of the segmented structure by right-clicking on
the “Segmentation” overlay and select “Display in 3D” (Fig. 16.5E).
14. Export the final model as a wavefront OBJ mesh file by selecting the “Export”
button to render high-quality images, or movies, using the Blender 3D modeling
software (version 2.69; Blender Foundation; http://www.blender.org)
(Fig. 16.5F).

3D analysis using NeuroMorph tools in Blender

15. Install the NeuroMorph tools into the Blender by selecting the “File”—“User
Preferences.” Under the “Addons” tab, select “Install from File” button and
locate where the addon files are stored. Select newly installed addon by
checking the right-hand box. Click “Save User Preferences” button and close
the “User Preferences” window. Repeat this procedure for all addon files.

Import tool
16. Under the “Scene” context panel, calibrate the scene by checking left-hand box
“Microns Per Pixel” and define the pixel size in microns (Fig. 16.6A).
17. Check the left-hand box “Use Remesh” to apply the octree remesh modifier to
imported object. This modifier reduces the number of vertices and memory
allocation, but still preserves the object shape. This can be tuned in “Remesh
resolution” field (higher number generates denser meshes and increases the
memory allocation).
18. To start import, click the “Import” button and locate the folder with stored OBJ
files. User can select multiple files which are then loaded and displayed in
Blender.
19. Imported objects are listed in the “Outliner” context panel. Select the object and
in the “Properties” context panel select “Modifier” tab. Further “Remesh”
modifier optimization can be tuned. By pressing “Apply” button the modifier is
applied.

Surface area/volume measurement

20. Define the type of structure to be measured in the “Measurement Tools” under
“Object Name” select the “Object type” pop-up menu with predefined
categories: spine, bouton, synapse, dendrite, axon, and other to define any new
category (Fig. 16.6B and C).
21. To measure the surface area and/or volume of the chosen structure, with the
right mouse button (RMB) select the object and switch to “Edit” mode using the
“Tab” key.
22. Activate “Vertex select” button and deactivate “Limit selection to visible”
button (both located at the bottom panel of the 3D view window).
356 CHAPTER 16 2-Photon Imaging and FIBSEM

A D E

3
B

2 G

1
4
I

1
2
3
4
5
6

FIGURE 16.6
NeuroMorph import, Measurement and Image Stack Interaction toolset for Blender. (A) The
mesh files created by Ilastik are imported into Blender using the Import tool that also
calibrates the size of the models into a micron scale. During the import process, an algorithm
(remesh modifier) is applied to reduce the number of vertices that comprises the mesh object
while also preserving the objects shape. (B) The Measurement tool is able to name the objects
being analyzed (here the names of neuronal structures are used: spine, bouton, synapse,
dendrite, axon, or other; indicated with yellow 1 (light gray in the print version), and in C).
After selecting the vertices of the chosen structure, the type of measurement is selected
(yellow 2 (light gray in the print version)). The results can be exported into a text file for further
analysis (yellow 4 (light gray in the print version)). (D) A model of dendrite with selected
 16.1 Method 357

23. Using lasso selection, by simultaneously pressing “Ctrl” key and the left mouse
button (LMB), select vertices of the chosen structure. Alternatively, “Border”
selection can be used by pressing “B” key and with LMB select the region.
Selected vertices are highlighted in orange (Fig. 16.6D).
24. In the “Measurement Tools” the user can select the type of measurement by
activating “Surface Area” or “Volume” buttons separately or the “Both” button
to perform both measurements simultaneously (Fig. 16.6B). Note that for
each measurement category a new child object is created and listed in the
“Outliner” context panel (Fig. 16.6E). For the surface area, an open mesh named
“surf” and for the volume a close mesh named “vol” are created. The
measurement values are displayed in the “Properties” context panel under the
“Object” tab and in the “Geometry Properties” section (Fig. 16.6F).

Length measurement
25. To measure the length, select the object with RMB and switch to “Edit” mode
“Tab” key.
26. Activate both “Vertex select” and “Limit selection to visible” buttons (both
located at the bottom panel of the 3D view window).
27. Select points (vertices) along the object to be measured, using RMB, adding
each point into the list by selecting “Add Point” button (or “Ctrl S” shortcut).
28. Activate “Multi-Segment Distance” button to measure the length from multiple
points. The new child object named “MS_dist” is created and listed in the
“Outliner” context panel. The length values are displayed in the “Properties”

FIGURE 16.6 —Cont’d vertices that define a dendritic spine (highlighted in orange (light gray
in the print version), left panel) is measured, creating a new mesh object (3) listed in the
Outliner window (E). Other structures (colorized in magenta (gray in the print version)) on this
dendrite model, and shown in the Object mode, can be measured (G), to form many mesh
objects (shown in the Outliner in (E)). (F) The result of these measurements is displayed in the
“Properties” context panel under the “Object” tab and in the “Geometry Properties” section.
(H) The Image Stack Interaction tool is able to display the EM image (size is defined in
microns; yellow 1 (light gray in the print version), and from the correct directory; yellow 2 (light
gray in the print version)), from which the model is derived, at the correct position in the model
(displayed in Edit mode after clicking on the button indicated with yellow 1 (light gray in the
print version)). User can also scroll through images of the entire stack (yellow 4 (light gray in
the print version)). The displayed image can also be used to retrieve the 3D model of any
structure (object) by selecting the points density (in pixels) and displaying the grid over the
image (5). Then by selecting any grid point(s), the corresponding object(s) is displayed at
selected vertex (6). (I) Retrieved objects based on the selection the grid points are displayed:
presynaptic bouton (in transparent green (light gray in the print version)), axonal
mitochondrion (in green (dark gray in the print version)), synapse (in red (dark gray in the
print version)), and dendritic mitochondrion (in magenta (dark gray in the print version)). The
initial dendrite is shown in transparent magenta (light gray in the print version).
358 CHAPTER 16 2-Photon Imaging and FIBSEM

context panel under the “Object” tab and in the “Geometry Properties” section
under Length of Curve. Alternatively, for shorter or relatively straight
segments, adding two points (start–end) to the list is sufficient for length
measurement by selecting “Shortest distance” button. The new child object
named “2p_dist” is created and listed in the “Outliner” panel (Fig. 16.6E).
29. When finished with length measurement, remove the points from the list with
“Clear points” button.

Export measurement results

30. Measurement results can be exported into a single CSV text file (.txt) for further
analysis with a spreadsheet program (e.g., Excel, Open Office, Numbers). By
activating the “Export Measurements” button in the “Measurement Tools”
under “Output Data” section, user specifies the destination folder and the file
name. All measurement results for all analyzed objects present in the blender
file are exported and written to the text file.
31. Alternatively, the measurement results for only selected object(s) can be
exported by activating the Export Selected” button.

Image Stack Interaction tool

32. This tool allows the interaction of the reconstructed model with the original EM
images (Fig. 16.6H and I). In the “Display Images from Stack” section under
“Image Stack Dimensions,” specify the reconstructed volume (in microns) size
in x, y, and z dimensions in appropriate fields.
33. In the “Source” section, locate and activate folder button (right-hand side) to
select the storage folder with the sequential images of the volume stack used for
the reconstruction.
34. To overlay the model with the image, select the object with RMB and switch to
edit mode pressing “Tab” key.
35. Select any vertex from the mesh and activate “Show image at Vertex” button.
The image from the volume stack at the position of selected vertex is placed
to the scene (called “Empty” with EM image as a texture). Note: user can
selected multiple vertices (RMB with “Shift” key) in which case for each
selected point, the image plane is added.
36. To see the EM image through the object, transparency can be changed by
selecting the object with RMB and activating the “Add Transparency” button in
the “Mesh Transparency” section. In the field “Alpha,” the opacity intensity
can be adjusted as well as the color can be changed from the “Color Palette”
button (next to “Alpha” button). To remove the transparency of the object,
select “Remove Transparency” button (Fig. 16.6H).
37. By selecting the image plane (or Empty), the user is able to scroll the image
plane up and down throughout the whole volume by activating the “Scroll
Through Image Stack” and by using mouse middle button. This shows each
 Acknowledgments 359

image from the original stack at its correct position relative to the model.
Pressing the “Esc” key deactivates this scrolling function.
38. The image plane can be used to display any additional object in the scene based
on the current image. Select the image plane (RMB) and define the grid
density (distance between grid points in pixels) in x- and y dimensions in the
“dx” and “dy” fields located under “Retrieve Object” from Image section.
39. By activating the “Display Grid” button, the mesh grid is overlayed over the
image plane. Select any point (RMB) or multiple points (RMB with “Shift” key).
40. To display the object based on grid point selection, activate the “Display Object
at Selected Vertex” button.

CONCLUSION
Recent years have seen significant advances in EM imaging methods for capturing
volumetric data on a large scale. In particular, the rise of block-face scanning ap-
proaches now offers many opportunities for automated serial imaging. These are
likely to improve, with more efficient detectors achieving better signal to noise ratios
giving faster imaging rates. More automated procedures that ensure consistent image
quality for longer periods will improve. All this will allow users to image more sam-
ples and allow these microscopes to be available to more users. Correlating light mi-
croscopy data with structural information from the electron microscope will,
therefore, become increasingly more prevalent.
The protocol presented here circumvents a number of the steps that have made
CLEM methods using transmission electron microscopy an exception rather than
the norm. In many places, it reduces considerably the need of highly skilled hands
to carry out such tasks as cutting, collecting, and imaging many hundreds of serial
sections. It also describes how to use a freely available set of software tools, useable
on any computer platform, to rebuild automatically, in the computer, the various tar-
geted structures, as well as carry out detailed morphometric analysis.
The computer vision research providing the segmentation and reconstruction
tools, described here is, however, dependent on large and varied images, as well
as ground truth segmentations to develop and test their algorithms. As more labora-
tories use these imaging methods, more datasets becomes available and better soft-
ware will lead to more opportunities to analyze larger image volumes, and reducing
the time needed to proofread and correct the results. This will inevitably lead to
greater insights into the structure and function of biological systems.

ACKNOWLEDGMENTS
This work was supported by the Swiss National Foundation Synergia project grant CRF
II313470/1 (G.W.K.), EU ERC MicroNano Project (P.F.) and project grant 31003A_135631
(A.H.).
360 CHAPTER 16 2-Photon Imaging and FIBSEM

REFERENCES
Allegra Mascaro, A. L., Cesare, P., Sacconi, L., Grasselli, G., Mandolesi, G., Maco, B., et al.
(2013). In vivo single branch axotomy induces GAP-43-dependent sprouting and synaptic
remodeling in cerebellar cortex. Proceedings of the National Academy of Sciences of the
United States of America, 110, 10824–10829.
Becker, C., Ali, K., Knott, G., & Fua, P. (2013). Learning context cues for synapse segmen-
tation. IEEE Transactions on Medical Imaging, 32, 1864–1877.
Bishop, D., Nikic, I., Brinkoetter, M., Knecht, S., Potz, S., Kerschensteiner, M., et al. (2011).
Near-infrared branding efficiently correlates light and electron microscopy. Nature
Methods, 8, 568–570.
Cane, M., Maco, B., Knott, G., & Holtmaat, A. (2014). The relationship between PSD-95 clus-
tering and spine stability in vivo. The Journal of Neuroscience, 34, 2075–2086.
Canty, A. J., Huang, L., Jackson, J. S., Little, G. E., Knott, G., Maco, B., et al. (2013). In-vivo
single neuron axotomy triggers axon regeneration to restore synaptic density in specific
cortical circuits. Nature Communications, 4, 2038.
Fiala, J. C., & Harris, K. M. (2001). Cylindrical diameters method for calibrating section thick-
ness in serial electron microscopy. Journal of Microscopy, 202, 468–472.
Grillo, F. W., Song, S., Teles-Grilo Ruivo, L. M., Huang, L., Gao, G., Knott, G. W., et al.
(2013). Increased axonal bouton dynamics in the aging mouse cortex. Proceedings of
the National Academy of Sciences of the United States of America, 110, E1514–E1523.
Holtmaat, A., Wilbrecht, L., Knott, G. W., Welker, E., & Svoboda, K. (2006). Experience-
dependent and cell-type-specific spine growth in the neocortex. Nature, 441, 979–983.
Knott, G. W., Holtmaat, A., Trachtenberg, J. T., Svoboda, K., & Welker, E. (2009). A protocol
for preparing GFP-labeled neurons previously imaged in vivo and in slice preparations for
light and electron microscopic analysis. Nature Protocols, 4, 1145–1156.
Knott, G. W., Holtmaat, A., Wilbrecht, L., Welker, E., & Svoboda, K. (2006). Spine growth
precedes synapse formation in the adult neocortex in vivo. Nature Neuroscience, 9,
1117–1124.
Kreshuk, A., Straehle, C. N., Sommer, C., Koethe, U., Cantoni, M., Knott, G., et al. (2011).
Automated detection and segmentation of synaptic contacts in nearly isotropic serial elec-
tron microscopy images. PLoS One, 6, e24899.
Lucchi, A., Smith, K., Achanta, R., Knott, G., & Fua, P. (2012). Supervoxel-based segmen-
tation of mitochondria in em image stacks with learned shape features. IEEE Transactions
on Medical Imaging, 31, 474–486.
Lucchi, A., Smith, K., Achanta, R., Lepetit, V., & Fua, P. (2010). A fully automated approach
to segmentation of irregularly shaped cellular structures in EM images. Medical Image
Computing and Computer-Assisted Intervention, 13, 463–471, International Conference
on Medical Image Computing and Computer-Assisted Intervention.
Maco, B., Holtmaat, A., Cantoni, M., Kreshuk, A., Straehle, C. N., Hamprecht, F. A., et al.
(2013). Correlative in vivo 2 photon and focused ion beam scanning electron microscopy
of cortical neurons. PLoS One, 8, e57405.
Mostany, R., Anstey, J. E., Crump, K. L., Maco, B., Knott, G., & Portera-Cailliau, C. (2013).
Altered synaptic dynamics during normal brain aging. Journal of Neuroscience, 33,
4094–4104.
Schultz, R. L., & Karlsson, U. (1965). Fixation ot the central nervous system for electron mi-
croscopy by aldehyde perfusion. II. Effect of osmolarity, pH of perfusate, and fixative con-
centration. Journal of Ultrastructure Research, 12, 187–206.
 References 361

Straehle, C. N., Köthe, U., Knott, G., & Hamprecht, F. A. (2011). Carving: Scalable interactive
segmentation of neural volume electron microscopy images. Medical Image Computing
and Computer-Assisted Intervention, 14, 653–660.
Tao-Cheng, J. H., Gallant, P. E., Brightman, M. W., Dosemeci, A., & Reese, T. S. (2007).
Structural changes at synapses after delayed perfusion fixation in different regions of
the mouse brain. The Journal of Comparative Neurology, 501, 731–740.
Trachtenberg, J. T., Chen, B. E., Knott, G. W., Feng, G., Sanes, J. R., Welker, E., et al. (2002).
Long-term in vivo imaging of experience-dependent synaptic plasticity in adult cortex.
Nature, 420, 788–794.
Zito, K., Parnas, D., Fetter, R. D., Isacoff, E. Y., & Goodman, C. S. (1999). Watching a synapse
grow: Noninvasive confocal imaging of synaptic growth in Drosophila. Neuron, 22,
719–729.