Automatic segmentation of high pressure frozen and freeze-substituted mouse retina nuclei from FIB-SEM tomograms

https://doi.org/10.1016/j.jsb.2016.10.005Get rights and content

Abstract

Focused Ion Beam milling combined with Scanning Electron Microscopy is a powerful tool to determine the 3-D organization of whole cells and tissue at an isotropic resolution of 3–5 nm. This opens the possibility to quantify several cellular parameters and to provide detailed phenotypic information in normal or disease states. Here we describe Biocomputing methods to extract in an automated way characteristic features of mouse rod photoreceptor nuclei such as the shape and the volume of the nucleus; the proportion of heterochromatin; the number, density and distribution of nuclear pore complexes (NPC). Values obtained on five nuclei show that the number of NPC (348 ± 8) is the most conserved feature. Nuclei in higher eukaryotes show large variations in size and rod nuclei are amongst the smallest reported (32 ± 3 μm3). Despite large species- and cell-type-specific variations in size, the density of NPC (about 15/μm2) is highly conserved.

Introduction

The internal organization of cells reflects the expressed part of their genome, determines their specialized role in different tissues, and notifies about their physiological or pathological state. Since the first electron micrograph of cells using transmission electron microscopy (TEM) in 1945 (Porter et al., 1945), electron microscopy (EM) methods have continually improved our appreciation of complex levels of cellular organization. The high spatial resolution of electron microscopes produces a wealth of information about the structure, distribution and interactions between cellular organelles. Early electron microscopy observations of cellular structures were essentially two-dimensional (2-D) and resulted in mostly descriptive information. Stereological methods were used to predict or estimate the size and distribution of cellular structures by extrapolating from 2-D measurements to the whole cellular volume (Weibel et al., 1966). To circumvent limitations due to the intricate shape of cellular structures which prevents straightforward extrapolation, new methods were developed to retrieve 3-D information from cellular samples. By sequentially tilting the specimen relative to the electron beam, TEM tomography produces a 3-D model of a cell section (Hoppe and Grill, 1977). This powerful method is however limited to 500 nm in the Z-depth by section thickness and the reconstruction of an entire eukaryotic cell can be very challenging (Noske et al., 2008). Serial thin sections and TEM imaging can solve the problem of the limited Z-depth but this technique is very labor intense and needs highly skilled experimenters (Porter and Blum, 1953, White et al., 1986, Bumbarger et al., 2013). The technological challenge for rapid 3-D imaging of large mammalian cells at nanometric scale was first met with the introduction of block face imaging with a scanning electron microscope (SEM). In this case, the block face of an embedded sample is imaged with sufficient resolution and depth discrimination using the back scattered electron (BSE) detectors. To obtain 3-D data, the block face needs to be repeatedly imaged, with the top slice removed between image acquisitions. One possibility is to remove the top slice with an ultramicrotome in the chamber of a SEM (serial block face, SFB-SEM). This method was introduced 30 years ago by Leighton who constructed a microtome for cutting sections inside the microscope chamber (Leighton, 1981). More recently, Denk and Horstmann showed that with a custom-designed microtome in a SEM, 3-D ultrastructural data can be obtained at the resolution and with a volume sufficient to follow local neuronal circuits (Denk and Horstmann, 2004). Currently, the Z-resolution of SBF-SEM of about 25 nm is limited by the minimal section thickness that can be cut with a diamond-knife (Friedrich et al., 2013, Kasthuri et al., 2007, Tapia et al., 2012). The removal of top slices in smaller increments can also be performed by focused gallium ion beam (FIB) milling (Heymann et al., 2006). Recently, the limits of FIB–SEM tomography were pushed to create high-resolution 5 × 5 × 5 nm isotropic voxel acquisition on 40 × 40 μm large image areas and over a depth of several tenths of micrometers (Narayan et al., 2014, Wei et al., 2012). The combination of focused ion beam (FIB) milling and block-face SEM imaging (FIB-SEM tomography) has become a powerful tool for 3-D investigation of cellular ultrastructure and constitutes currently one of the most fast-growing methods to obtain undistorted volume information (Bosch et al., 2015, Cretoiu et al., 2015, De Winter et al., 2009, Hekking et al., 2009, Heymann et al., 2009, Knott et al., 2008, Merchán-Pérez et al., 2009). This new imaging technology opens the possibility to measure morphological parameters on discrete ultrastructural feature. Along with these large 3-D image stacks come many computational challenges to achieve automated 3-D object recognition, annotation, data integration, representation and exploitation. Segmentation is central to these tasks and corresponds to the process of partitioning the 3-D images into multiple structurally defined regions to produce a more meaningful representation suitable for quantitative analysis. This process can be used to delineate the boundaries between cellular compartments or between organelles and cytosol so that quantitative information such as the size, distribution, surface morphology can be extracted from these delineated regions.

Many segmentation techniques have been proposed to target nature scene and medical images. These techniques usually group nearby pixels into regions by optimizing a target function that prefers pixels in the same group to have similar intensity levels and at the same time expects a small number of groups (Arbelaez et al., 2011, Van den Bergh et al., 2014). In spite of many successful applications, these techniques are not suitable for electron microscopy images because of their low signal-to-noise ratio, low contrast and complex image contents. For example, in the presence of noise the performance of watershed transform-based methods (Couprie et al., 2011) decreases rapidly since noise changes the topological landscape of the intensity function representing the image contents and thus breaks biologically-meaningful regions into many small fragments. Recently, some research groups have developed algorithms and software tools that can aid the segmentation of electron microscopy images automatically (Nunez-Iglesias et al., 2014, Schindelin et al., 2012). For boundary segmentation, membrane signals are usually enhanced by filtering and then traced by thresholding (Martinez-Sanchez et al., 2014). For region segmentation, a superpixel-based method is often used to over-segment the image and then a split-and-merge strategy is used for region grouping (Jones et al., 2015). A classifier trained with human annotated data can also be used for direct pixel classification (Sommer et al., 2011). These methods were mostly developed for segmentation of brain tissue images with 25 nm resolution in the Z-direction. However, most of the developed segmentation algorithms work only with 2-D images and much human effort is still required to correct the automated segmentation results.

Here we have developed algorithms and software tools in a high-throughput computational workflow covering image pre-processing, semi-automatic segmentation of cell nuclei, and extraction of quantitative data from 3-D FIB-SEM images of murine rod photoreceptor cell nuclei. How our genetic information is organized in the cell nucleus and how this organization affects nuclear function are major questions in the chromatin field which remain largely unanswered. In eukaryotes, the genetic information is packaged through a hierarchy of folding events. At the first level DNA is wrapped around an histone octamer to form the nucleosome core particle whose atomic structure is defined (Luger et al., 1997). The fundamental repeated element of chromatin, the nucleosome, is composed of the core particle flanked by linker DNA which connects successive nucleosomes and variable amounts of linker histone H1. Purified chromatin appears as an extended 11-nm fiber formed by a linear beads-on-a-string nucleosomal array that compacts into 30 nm fibers in physiological ionic strength and in the presence of histone H1 (Oudet et al., 1975, Thoma et al., 1979). Direct electron microscopy imaging of nuclear sections described highly compact electron dense heterochromatin (HC) compartments and more extended euchromatin (EC) territories but has provided little information on chromatin organization beyond the nucleosomal level and in particular has not confirmed the 30-nm fibers as the fundamental in vivo secondary structure of chromatin (Bouchet-Marquis et al., 2005, Horowitz-Scherer and Woodcock, 2005). However, recent publications provide evidence that 30 nm chromatin fibers may existence in highly condensed chromatin states (Kizilyaprak et al., 2010). The chromatin organization in intact nuclei is thus still a matter of debate.

The tools developed here extend previous quantification efforts of nuclear structures (Rouquette et al., 2009) and assign automatically the extent of the nuclear envelope and measure the size of the nuclei as well as its elongation. The nuclear envelope is perforated by nuclear pores that regulate the exchanges between the nucleus and the cytoplasm. The number and distribution of the pores are determined automatically without using correlation methods. Finally the highly condensed heterochromatin compartment is identified automatically thus opening the possibility to model the rheology of the nucleus within its numerous channels or the diffusion of transcription factors within the nuclei.

Section snippets

Data collection

To address the quantification of ultrastructural features and particularly the very sensitive nuclei considerations have to be given concerning sample preparation in order to image cells in their closest to native state. In this study, we used high-pressure freezing (HPF) and freeze-substitution (FS) to prepare 4-days old mice retina samples. High-pressure freezing is a method of choice to obtain optimal structural preservation of bulk biological samples with minimal structural rearrangements (

Image pretreatment before segmentation

The FIB-SEM technology provides 3-D images of cells and tissues with isotropic resolution, thus opening the possibility to extract quantitative information from the collected data. Following the aim of analyzing nuclear images in a semi-automated way, image corrections were found necessary for accurate intensity-based image segmentation and special care was taken to implement methods that reduce noise without loss of spatial resolution and contrast. Global variations in image intensity which

Retina dissection

Four days old C57/BL6 wild-type mice were sacrificed according to institutional ethical guidelines and their eyes were rapidly enucleated and placed in Artificial CerebroSpinal Fluid (ACSF) buffer (126 mM NaCl, 2.5 mM KCl, 1.2 mM MgCl2, 2.4 mM CaCl2, 1.2 mM NaH2PO4, 18 mM NaHC03, 11 mM Glucose, PH 7.4, equilibrated with 95% O2 and 5% CO2) for dissection. The retinas were gently separated from the pigment epithelium and the sclera after removal of the lens, iris and vitreous humor. The dissection was

Competing interests

The authors declare no competing financial interests.

Author contributions

V.T.H performed the image analysis of the FIB/SEM data, C.K. acquired the FIB/SEM data, D.S. prepared the samples for electron microscopy, P.S. and B.M.H. wrote the manuscript. All authors reviewed the manuscript.

Funding

This work was supported by the Fondation pour la Recherche Médicale, the Institut National de la Santé et de la Recherche Médicale, the Centre National pour la Recherche Scientifique, the Strasbourg University, the Association pour la Recherche sur le Cancer and the Agence Nationale pour la Recherche (CHROMACT), the French Infrastructure for Integrated Structural Biology (FRISBI) [ANR-10-INSB-05-01] and INSTRUCT as part of the European Strategy Forum on Research Infrastructures (ESFRI). By the

Acknowledgements

We thank Jonathan Michalon for setting up a competitive computing environment, Dr. Chengge Jiao for suggestion about the optimized sample geometry and Céline Ziegler-Birling for genotyping and taking care of the animals.

References (79)

  • A. Martinez-Sanchez et al.

    Robust membrane detection based on tensor voting for electron tomography

    J. Struct. Biol.

    (2014)
  • R. Medina-Carnicer et al.

    On candidates selection for hysteresis thresholds in edge detection

    Pattern Recognit.

    (2009)
  • K. Narayan et al.

    Multi-resolution correlative focused ion beam scanning electron microscopy: applications to cell biology

    J. Struct. Biol.

    (2014)
  • A.B. Noske et al.

    Expedited approaches to whole cell electron tomography and organelle mark-up in situ in high-pressure frozen pancreatic islets

    J. Struct. Biol.

    (2008)
  • P. Oudet et al.

    Electron microscopic and biochemical evidence that chromatin structure is a repeating unit

    Cell

    (1975)
  • M.P. Scheffer et al.

    Nucleosomes stacked with aligned dyad axes are found in native compact chromatin in vitro

    J. Struct. Biol.

    (2012)
  • I. Solovei et al.

    Nuclear architecture of rod photoreceptor cells adapts to vision in mammalian evolution

    Cell

    (2009)
  • E.L. West et al.

    Pharmacological disruption of the outer limiting membrane leads to increased retinal integration of transplanted photoreceptor precursors

    Exp. Eye Res.

    (2008)
  • P. Arbelaez et al.

    Contour detection and hierarchical image segmentation

    IEEE Trans. Pattern Anal. Mach. Intell.

    (2011)
  • G. Babaloukas et al.

    Evaluation of three methods for retrospective correction of vignetting on medical microscopy images utilizing two open source software tools

    J. Microsc.

    (2011)
  • D. Barash

    Fundamental relationship between bilateral filtering, adaptive smoothing, and the nonlinear diffusion equation

    IEEE Trans. Pattern Anal. Mach. Intell.

    (2002)
  • C. Bosch et al.

    FIB/SEM technology and high-throughput 3D reconstruction of dendritic spines and synapses in GFP-labeled adult-generated neurons

    Front. Neuroanat.

    (2015)
  • C. Bouchet-Marquis et al.

    Cryoelectron microscopy of vitrified sections: a new challenge for the analysis of functional nuclear architecture

    Histochem. Cell Biol.

    (2005)
  • T.F. Chan et al.

    Active contours without edges

    IEEE Trans. Image Process

    (2001)
  • C. Couprie et al.

    Power watershed: a unifying graph-based optimization framework

    IEEE Trans. Pattern Anal. Mach. Intell.

    (2011)
  • D. Cretoiu et al.

    FIB-SEM tomography of human skin telocytes and their extracellular vesicles

    J. Cell Mol. Med.

    (2015)
  • K. Dabov et al.

    Image denoising by sparse 3-D transform-domain collaborative filtering

    IEEE Trans. Image Process.

    (2007)
  • W. Denk et al.

    Serial block-face scanning electron microscopy to reconstruct three-dimensional tissue nanostructure

    PLoS Biol.

    (2004)
  • D.A.M. De Winter et al.

    Tomography of insulating biological and geological materials using focused ion beam (FIB) sectioning and low-kV BSE imaging

    J. Microsc.

    (2009)
  • R.W. Friedrich et al.

    Analyzing the structure and function of neuronal circuits in zebrafish

    Front. Neural Circuits

    (2013)
  • L.M. Garcia-Segura et al.

    Distribution of nuclear pores and chromatin organization in neurons and glial cells of the rat cerebellar cortex

    J. Comp. Neurol.

    (1989)
  • R.C. Gonzalez et al.

    Digital Image Processing

    (2007)
  • J.A. Hartigan et al.

    Algorithm AS 136: A k-means clustering algorithm

    J. R. Stat. Soc. Ser. C Appl. Stat.

    (1979)
  • T.J. Hastie et al.

    The Elements of Statistical Learning: Data Mining, Inference, and Prediction

    (2009)
  • E. Hatch et al.

    Breaching the nuclear envelope in development and disease

    J. Cell Biol.

    (2014)
  • L.H.P. Hekking et al.

    Focused ion beam-scanning electron microscope: exploring large volumes of atherosclerotic tissue

    J. Microsc.

    (2009)
  • R.A. Horowitz-Scherer et al.

    Organization of interphase chromatin

    Chromosoma

    (2005)
  • B.M. Humbel

    Freeze-substitution

  • B.M. Humbel et al.

    Improved structural preservation by combining freeze substitution and low temperature embedding

    Beitr. Elektronenmikrosk. Direktabb Oberfl

    (1983)
  • Cited by (0)

    View full text