Limits...
Correlative in vivo 2 photon and focused ion beam scanning electron microscopy of cortical neurons.

Maco B, Holtmaat A, Cantoni M, Kreshuk A, Straehle CN, Hamprecht FA, Knott GW - PLoS ONE (2013)

Bottom Line: Correlating in vivo imaging of neurons and their synaptic connections with electron microscopy combines dynamic and ultrastructural information.These neurites are then identified and reconstructed automatically from the image series using the latest segmentation algorithms.The fast and reliable imaging and reconstruction technique avoids any specific labeling to identify the features of interest in the electron microscope, and optimises their preservation and staining for 3D analysis.

View Article: PubMed Central - PubMed

Affiliation: BioEM Facility, Centre of Electron Microscopy, EPFL, Lausanne, Switzerland.

ABSTRACT
Correlating in vivo imaging of neurons and their synaptic connections with electron microscopy combines dynamic and ultrastructural information. Here we describe a semi-automated technique whereby volumes of brain tissue containing axons and dendrites, previously studied in vivo, are subsequently imaged in three dimensions with focused ion beam scanning electron microcopy. These neurites are then identified and reconstructed automatically from the image series using the latest segmentation algorithms. The fast and reliable imaging and reconstruction technique avoids any specific labeling to identify the features of interest in the electron microscope, and optimises their preservation and staining for 3D analysis.

Show MeSH
Interactive segmentation of neurites.A, The ilastik software allows users to select regions for segmentation (yellow box), and compare the generated 3D model (B) with the in vivo image (C) of the neurite of interest. A GFP dendrite from a layer 5 pyramidal neuron is shown in C. The method can also be used with other fluorescent markers, such as tdTomato (D), here expressed in GABAergic axons and dendrites in a different animal. When the density of neurites (E) is high the neurites can still be distinguished. Panel E, shows the reconstructed axons (red) and dendrites (grey) that are shown in the in vivo image in D. Scale bar in C and D is 5 µm.
© Copyright Policy
Related In: Results  -  Collection


getmorefigures.php?uid=PMC3585404&req=5

pone-0057405-g002: Interactive segmentation of neurites.A, The ilastik software allows users to select regions for segmentation (yellow box), and compare the generated 3D model (B) with the in vivo image (C) of the neurite of interest. A GFP dendrite from a layer 5 pyramidal neuron is shown in C. The method can also be used with other fluorescent markers, such as tdTomato (D), here expressed in GABAergic axons and dendrites in a different animal. When the density of neurites (E) is high the neurites can still be distinguished. Panel E, shows the reconstructed axons (red) and dendrites (grey) that are shown in the in vivo image in D. Scale bar in C and D is 5 µm.

Mentions: This small z depth between images ensures small changes in structural information between each one, giving the best opportunity for 3D segmentation algorithms to reconstruct the neurites. We used the interactive segmentation tool [14] inside the ilastik software (www.ilastik.org) to reconstruct neurites from the series. Given user input, this tool performs a weighted watershed segmentation on a superpixel graph of the 3D stack. Specifically, the user needs to mark the inside (object) and the outside (background) parts of the area of interest (Fig. 2A). Provided that its boundaries, ie. the membranes, are well stained, only a few (2–8) seeds are required to fully reconstruct an object such as the neurite shown here (Fig. 2). This number can be lower if the user is only interested in the overall topology and direction of the neurite, rather than capturing all protrusions, like spines from the dendritic shaft. The one-time construction of a superpixel graph renders the segmentation process fully interactive and allows the user to immediately assess the results in 2D (Fig. 2A) and 3D (Fig. 2B). These properties make ilastik especially suitable for quickly locating the labeled neurites in the EM image stack: only a few clicks are needed to establish if a given object continues in the same direction or branches in the same way as the labeled neurite in the in vivo 2 photon image (GFP labeled neurites, Fig. 2C; and tdTomato labeled neurites, Fig. 2D). After the labeled neurite candidates are found, their segmentation can be refined by additional object and background seeds, correcting the areas where segments bled into neighboring objects or were cut short by a heavily stained internal membrane. The total time for the basic reconstruction of the dendrites and axons shown in Fig. 2E was approximately 2 hours (stack containing of 1505 images). This is a fraction of the time that would be required if this task was to be undertaken by manually segmenting each feature in the serial images.


Correlative in vivo 2 photon and focused ion beam scanning electron microscopy of cortical neurons.

Maco B, Holtmaat A, Cantoni M, Kreshuk A, Straehle CN, Hamprecht FA, Knott GW - PLoS ONE (2013)

Interactive segmentation of neurites.A, The ilastik software allows users to select regions for segmentation (yellow box), and compare the generated 3D model (B) with the in vivo image (C) of the neurite of interest. A GFP dendrite from a layer 5 pyramidal neuron is shown in C. The method can also be used with other fluorescent markers, such as tdTomato (D), here expressed in GABAergic axons and dendrites in a different animal. When the density of neurites (E) is high the neurites can still be distinguished. Panel E, shows the reconstructed axons (red) and dendrites (grey) that are shown in the in vivo image in D. Scale bar in C and D is 5 µm.
© Copyright Policy
Related In: Results  -  Collection

Show All Figures
getmorefigures.php?uid=PMC3585404&req=5

pone-0057405-g002: Interactive segmentation of neurites.A, The ilastik software allows users to select regions for segmentation (yellow box), and compare the generated 3D model (B) with the in vivo image (C) of the neurite of interest. A GFP dendrite from a layer 5 pyramidal neuron is shown in C. The method can also be used with other fluorescent markers, such as tdTomato (D), here expressed in GABAergic axons and dendrites in a different animal. When the density of neurites (E) is high the neurites can still be distinguished. Panel E, shows the reconstructed axons (red) and dendrites (grey) that are shown in the in vivo image in D. Scale bar in C and D is 5 µm.
Mentions: This small z depth between images ensures small changes in structural information between each one, giving the best opportunity for 3D segmentation algorithms to reconstruct the neurites. We used the interactive segmentation tool [14] inside the ilastik software (www.ilastik.org) to reconstruct neurites from the series. Given user input, this tool performs a weighted watershed segmentation on a superpixel graph of the 3D stack. Specifically, the user needs to mark the inside (object) and the outside (background) parts of the area of interest (Fig. 2A). Provided that its boundaries, ie. the membranes, are well stained, only a few (2–8) seeds are required to fully reconstruct an object such as the neurite shown here (Fig. 2). This number can be lower if the user is only interested in the overall topology and direction of the neurite, rather than capturing all protrusions, like spines from the dendritic shaft. The one-time construction of a superpixel graph renders the segmentation process fully interactive and allows the user to immediately assess the results in 2D (Fig. 2A) and 3D (Fig. 2B). These properties make ilastik especially suitable for quickly locating the labeled neurites in the EM image stack: only a few clicks are needed to establish if a given object continues in the same direction or branches in the same way as the labeled neurite in the in vivo 2 photon image (GFP labeled neurites, Fig. 2C; and tdTomato labeled neurites, Fig. 2D). After the labeled neurite candidates are found, their segmentation can be refined by additional object and background seeds, correcting the areas where segments bled into neighboring objects or were cut short by a heavily stained internal membrane. The total time for the basic reconstruction of the dendrites and axons shown in Fig. 2E was approximately 2 hours (stack containing of 1505 images). This is a fraction of the time that would be required if this task was to be undertaken by manually segmenting each feature in the serial images.

Bottom Line: Correlating in vivo imaging of neurons and their synaptic connections with electron microscopy combines dynamic and ultrastructural information.These neurites are then identified and reconstructed automatically from the image series using the latest segmentation algorithms.The fast and reliable imaging and reconstruction technique avoids any specific labeling to identify the features of interest in the electron microscope, and optimises their preservation and staining for 3D analysis.

View Article: PubMed Central - PubMed

Affiliation: BioEM Facility, Centre of Electron Microscopy, EPFL, Lausanne, Switzerland.

ABSTRACT
Correlating in vivo imaging of neurons and their synaptic connections with electron microscopy combines dynamic and ultrastructural information. Here we describe a semi-automated technique whereby volumes of brain tissue containing axons and dendrites, previously studied in vivo, are subsequently imaged in three dimensions with focused ion beam scanning electron microcopy. These neurites are then identified and reconstructed automatically from the image series using the latest segmentation algorithms. The fast and reliable imaging and reconstruction technique avoids any specific labeling to identify the features of interest in the electron microscope, and optimises their preservation and staining for 3D analysis.

Show MeSH