for3d: full organ reconstruction in 3d, an automatized ... · for3d: full organ reconstruction in...

12
See discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/276067524 For3D: Full Organ Reconstruction in 3D, an Automatized Tool for Deciphering the Complexity of Lymphoid Organs Article in Journal of immunological methods · May 2015 DOI: 10.1016/j.jim.2015.04.019 · Source: PubMed CITATION 1 READS 36 5 authors, including: Arnauld Sergé Aix-Marseille Université 34 PUBLICATIONS 772 CITATIONS SEE PROFILE Beat A Imhof University of Geneva 306 PUBLICATIONS 10,967 CITATIONS SEE PROFILE Magali Irla Centre d'Immunologie de Marseille-Luminy 55 PUBLICATIONS 489 CITATIONS SEE PROFILE All content following this page was uploaded by Arnauld Sergé on 08 April 2016. The user has requested enhancement of the downloaded file. All in-text references underlined in blue are linked to publications on ResearchGate, letting you access and read them immediately.

Upload: vandat

Post on 15-Aug-2019

223 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: For3D: Full Organ Reconstruction in 3D, an Automatized ... · For3D: Full organ reconstruction in 3D, an automatized tool for deciphering the complexity of lymphoid organs Arnauld

Seediscussions,stats,andauthorprofilesforthispublicationat:https://www.researchgate.net/publication/276067524

For3D:FullOrganReconstructionin3D,anAutomatizedToolforDecipheringtheComplexityofLymphoidOrgans

ArticleinJournalofimmunologicalmethods·May2015

DOI:10.1016/j.jim.2015.04.019·Source:PubMed

CITATION

1

READS

36

5authors,including:

ArnauldSergé

Aix-MarseilleUniversité

34PUBLICATIONS772CITATIONS

SEEPROFILE

BeatAImhof

UniversityofGeneva

306PUBLICATIONS10,967CITATIONS

SEEPROFILE

MagaliIrla

Centred'ImmunologiedeMarseille-Luminy

55PUBLICATIONS489CITATIONS

SEEPROFILE

AllcontentfollowingthispagewasuploadedbyArnauldSergéon08April2016.

Theuserhasrequestedenhancementofthedownloadedfile.Allin-textreferencesunderlinedinblue

arelinkedtopublicationsonResearchGate,lettingyouaccessandreadthemimmediately.

Page 2: For3D: Full Organ Reconstruction in 3D, an Automatized ... · For3D: Full organ reconstruction in 3D, an automatized tool for deciphering the complexity of lymphoid organs Arnauld

Journal of Immunological Methods 424 (2015) 32–42

Contents lists available at ScienceDirect

Journal of Immunological Methods

j ourna l homepage: www.e lsev ie r .com/ locate / j im

For3D: Full organ reconstruction in 3D, an automatized tool fordeciphering the complexity of lymphoid organs

Arnauld Sergé a,⁎, Anne-Laure Bailly a, Michel Aurrand-Lions a, Beat A. Imhof b, Magali Irla c,⁎⁎a Centre de Recherche en Cancérologie de Marseille, Institut Paoli-Calmettes, Inserm U1068, CNRS UMR7258, Aix-Marseille Université UM105, Franceb Centre Médical Universitaire, Département Pathologie et Immunologie, Faculté de Médecine, Université de Genève, 1, rue Michel Servet, 1211 Genève, Switzerlandc Centre d'Immunologie de Marseille-Luminy, Inserm U1104, CNRS UMR7280, Aix-Marseille Université UM2, France

Abbreviations: 3D, three-dimensions; 2D, twOrgan Reconstruction in Three Dimensions; IF,immunohistochemistry; K, keratin; LN, lymph node.⁎ Correspondence to: A. Sergé, Centre de Recherche

(CRCM), 27 bd Leï Roure BP 30059, 13273 Marseille Cede72 93.⁎⁎ Correspondence to: M. Irla, Centre d'ImmunologieCampus de Luminy case 906, 13288 Marseille Cedex 09, F

E-mail addresses: [email protected] (A. Sergé

http://dx.doi.org/10.1016/j.jim.2015.04.0190022-1759/© 2015 Elsevier B.V. All rights reserved.

a b s t r a c t

a r t i c l e i n f o

Article history:Received 10 March 2015Received in revised form 28 April 2015Accepted 29 April 2015Available online 5 May 2015

Keywords:ThymusLymph nodePrimary lymphoid tumorOrgan topologyImage processing3D reconstruction

To decipher the complex topology of lymphoid structures, we developed an automated process called Full OrganReconstruction in 3D (For3D). A dedicated image-processing pipeline is applied to entire collections ofimmunolabeled serial sections, acquired with a slide-scanning microscope. This method is automated, flexibleand readily applicable in two days to frozen or paraffin-embedded organs stained by fluorescence or brightfieldimmunohistochemistry. 3D-reconstructed organs can be visualized, rotated and analyzed to quantify substruc-tures of interest. Usefulness of For3D is exemplified here through topological analysis of severalmouse lymphoidorgans exhibiting a complex organization: (i) the thymus, composed of two compartments, a medulla intricatelyimbricated into a surrounding cortex, (ii) lymph nodes, also highly compartmentalized into cortex, paracortexand medulla and (iii) the vascularization of an EG7 primary thymoma. This open-source algorithm, based onImageJ and Matlab scripts, offers a user-friendly interface and is widely applicable to any organ or tissue,hence readily adaptable to a broad range of biomedical samples.

© 2015 Elsevier B.V. All rights reserved.

1. Introduction

Lymphoid organs exhibit a complex topology, which is tightly associ-atedwith their primary function. The complexity of lymphoid tissueswasoriginally unraveled by widely used histological techniques, such as tis-sue counterstaining and immunolabeling, which allows the detection ofspecific molecular or cellular features. Visualizing the three-dimensional(3D) organization of organs is critical for understanding their structureand function. However, current histological methods used for examiningtopological organization remain largely inadequate because they rely onstaining of tissue sections and visualization of structures in two-dimensions (2D). Rigorous volume quantification can be hardly estimat-ed based solely on 2D observations. Two-photon imaging technology hasrecently allowed visualization of the 3D organization of some organsusing ex vivo organ explants or in vivo surgically exposed organs(Germain et al., 2006; Le Borgne et al., 2009). However, due to technicallimitations in tissue penetration, determining the 3D topology of entire

o-dimensions; For3D, Fullimmunofluorescence; IHC,

en Cancérologie de Marseillex 09, France, Tel.: +33 4 86 97

de Marseille-Luminy (CIML),rance. Tel.: +33 4 91 26 94 43.), [email protected] (M. Irla).

organs often remains largely inaccessible to bi-photon imaging (Niesnerand Hauser, 2011). Our knowledge of specific anatomical structures in3D, their number and relative organization, aswell as their respective vol-ume is thus limited. Recently, light sheet fluorescence microscopy wasdeveloped as an appealing alternative for 3D observations. The sampleis rendered transparent and labeled with fluorescent probes. A planar il-lumination beam is used to excite the sample from the side, and imagesare acquired while scanning the sample over its height, using a secondobjective, positioned orthogonally to the one used for illumination(Voie et al., 1993; Erturk et al., 2012; Gao et al., 2014; Reynaud et al.,2015). However, this recent method, requiring specific microscopesetup and sample preparation, is not yet frequently available. Further-more, antibody penetration inside thick samples may be limited.

To circumvent these limitations, we have developed an automatizedmethod named For3D, for Full Organ Reconstruction in 3D, allowing com-putational 3D rendering of organs from the entire collection ofimmunolabeled serial sections. We initially developed this approach forreconstructing the mouse thymus in 3D, which is the main site of T-cellproduction. Histological analyses based on 2D thymic sections have large-ly established that the thymus is anatomically divided into two main re-gions: an outer region called the cortex and an inner region called themedulla (Pearse, 2006). The cortex supports early stages of T-cell differ-entiation,whereas themedulla supports the late stages of T-cell selection,consistingmainly in the deletion of potentially hazardous T cells and thusin the generation of a self-tolerant T-cell repertoire (Palmer, 2003;Takahama, 2006; Anderson et al., 2007). These distinct steps of thymocyte

Page 3: For3D: Full Organ Reconstruction in 3D, an Automatized ... · For3D: Full organ reconstruction in 3D, an automatized tool for deciphering the complexity of lymphoid organs Arnauld

Organ, e.g. thymus, LN or tumor

SectioningFrozen or paraffin

IF or IHC staining

Image registering

Image acquisitionAutomated scanner

A

B

C

D

E

33A. Sergé et al. / Journal of Immunological Methods 424 (2015) 32–42

differentiation are controlled by the thymic epithelium. Cortical thymicepithelial cells mediate positive selection of CD4+ and CD8+ thymocytes,whereas medullary thymic epithelial cells are crucially involved in purg-ing the T-cell repertoire of auto-reactive T cells (Derbinski et al., 2001;Irla et al., 2010; Anderson and Takahama, 2012). Despite this extendedknowledge on T-cell differentiation, the 3D corticomedullary organizationremained largely elusive. We recently addressed this issue in health andimmunodeficiency (Irla et al., 2013), and showed as a proof-of-conceptof the For3Dapproach that themousemedulla is composed of a large cen-tral compartment, surrounded by tens of smaller medullae.

Since this 3D rendering has broad applications, we further devel-oped the method, as a comprehensive package, from sample prepara-tion to data collection and analysis. In addition to the mouse thymus,stained by immunofluorescence (IF), we show here that For3D can beapplied to other lymphoid organs, such as mesenteric lymph nodes(LNs) stained by brightfield immunohistochemistry (IHC). LNs consti-tute privileged sites for initiating adaptive immune responses throughantigen presentation. This secondary lymphoid organ is histologicallyseparated into twomain regions, the cortex and themedulla. The cortexis further divided into the paracortex, enriched in T cells, and follicles,enriched in B cells (von Andrian and Mempel, 2003). This architectureallows T and B cells to interact with antigen-presenting cells (Willard-Mack, 2006). Therefore, documenting the 3D organization of LNs is ofparticular interest for better apprehending the initiation of the adaptiveimmune response.

We further extended For3D to other applications such as the vascu-larization of primary tumors stained by IF. Solid tumors promote neo-vascularization by secreting factors such as VEGF (De Bock et al.,2013), thereby inducing the emergence of new vessels, initially at theperiphery, then gradually inside the tumor. This favors fetching tumorcells with oxygen and nutrients, thereby preventing necrosis. Ultimate-ly, vascularization even permits dissemination of tumor cells, potential-ly leading to metastasis. For these reasons, tumor angiogenesis is thetarget of several anti-cancer therapies aimed at blocking tumor growth(Welti et al., 2013). A detailed characterization of the complex architec-ture of tumor vascularization in 3D could be highly informative to fur-ther our understanding of tumor growth.

For3D is a user-friendly automated method for reconstructing the 3Darchitecture of organs of interest from tissue sections immunolabeledwith multiple antibodies. In addition to IF labeling, the For3D processcan also be applied to alternative labeling methods, such as brightfieldIHC or hematoxylin–eosin–safran imaged by transmitted light. We de-scribe here the For3D reconstruction process, which combines a collec-tion of serial immunolabeled sections with a dedicated image analysissoftware program. Matlab codes, handled by a user-friendly interfaceand fully documented in a user guide, are freely available to academicreaders upon request.

3DrenderingFilteringQuantificationFractal analysis

F

Visualization of the 3D organization

Fig. 1.Major steps of the For3D process. (A) A frozen or paraffin embedded tissue is entire-

2. Materials and methods

2.1. Animals

Mice were on a C57BL/6 background and were housed under specificpathogen-free conditions. For thymi and LNs, mice were sacrificed be-tween5 to 8weeks of age andorgans of interestwere carefully harvested.All procedures involving animals have been performed in accordancewith the institutional and ethical guidelines.

ly sectioned. (B) All sections are labeled by either immunofluorescence (IF) or brightfieldimmunohistochemistry (IHC) to detect structures or cells of interest. To visualize the en-tire tissue, sections are also counterstained with either the nucleic acid stain DAPI for IF,or hematoxylin for IHC. (C) Sections are imaged by either epifluorescence or brightfieldmicroscopy using an automated scanner. (D) Images are registered with respect to eachother using the StackReg plugin (Thevenaz et al., 1998) from the freeware ImageJ.(E) Images are further processed for equalization and filtering using the Matlab home-made function “rendering_3D”, leading to a 3D volumetric view of the organ (F). Furtheranalyses permit quantification of the resulting 3D structures, typically by object numera-tion andvolumetricmeasures, using 3D analysis software such as ImageJ,Matlab or Imaris.

2.2. Induction of EG7 primary tumors

Onemillion EG7 cells (ATCC, CRL2113) were injected subcutaneouslyinto the shaved back flank of C57BL/6 mice. Tumor size was monitoreddaily. Two weeks after injection, mice were sacrificed and tumors wereharvested.

2.3. Organ sectioning

An overview of the For3D process is illustrated in Fig. 1. After dissec-tion of the organ, surrounding fat tissue was removed to avoid any in-terference with the 3D reconstruction process. For the thymus andtumors, frozen sections were performed, while for the LN, paraffin sec-tions were performed. For frozen tissues, inclusion was performed byusing a mold (Leica) prefilled with OCT compound (Sakura). The speci-menwas gently immergedwith a longitudinal orientation in themiddledepth of themold and the inclusion was quick-frozen in liquid nitrogenor alternatively stored at−80 °C for at least 5 h to several months, tak-ing care that the tissue does not change orientation during this step. TheOCT block was mounted into a cryostat (CM3050 Sp, Leica) with achamber pre-cooled at −20 °C. The section thickness was chosen, as acompromise, to be large enough to lead to a reasonably low sectionnumber aswell as small enough to allow efficient labeling and adequateresolution in depth. 20-μm thick serial sections (Fig. 1A) were collectedon Superfrost plus glass slides (Fischer Scientific). Generally, five to sixsections were collected per slide. At this stage, air-dried slides wereused directly for immunostaining or stored at −80 °C for severalmonths.

For lymph nodes, tissues were fixed in 1% paraformaldehyde formaximum 4 h at room temperature. To avoid over-fixation, tissues

Page 4: For3D: Full Organ Reconstruction in 3D, an Automatized ... · For3D: Full organ reconstruction in 3D, an automatized tool for deciphering the complexity of lymphoid organs Arnauld

34 A. Sergé et al. / Journal of Immunological Methods 424 (2015) 32–42

were stored at 4 °C in 30% sucrose. For the subsequent steps, specimenswere automatically treated by a tissue processor (ASP300 Leica). Su-crose was removed in a 10% formaldehyde/alcohol bath during 5 minat 37 °C. Tissues were dehydrated through a series of absolute ethanolbaths, cleared in a 1-h histolemon bath and infiltrated with paraffinwax baths at 58 °C. Wax-infiltrated samples were then manually em-bedded using metallic molds. Paraffin solidification was obtained byplacing molds on a cold surface. Embedded tissues were sectionedusing a microtome equipped with Section Transfer System (HM340EMicrom). Ribbons of 5-μm thick serial sections were transferred fromthe laminar flow to glass slides. 5 to 7 serial sections were collectedper slide.

2.4. Section immunostaining

For the thymus and tumors, the entire procedure of section immu-nostaining was performed in a humidified chamber at room tempera-ture. Sections were fixed in 2% paraformaldehyde (Sigma Aldrich)during 15 min, and carefully washed three times for 5 min with 0.1 MTris buffer (Invitrogen) pH 7.4. Sections were permeabilized for10 min in 0.1 M Tris buffer pH 7.4, 0.02% Triton X-100 (Sigma Aldrich).Before incubation with primary antibodies, sections were incubated for10min in saturation buffer composed of 0.1M Tris buffer pH 7.4, 3% BSA(Axday), 0.01% Triton X-100.

Thymic sections were stained for 30 min with primary antibodies,rabbit anti-keratin 14 (K14, clone AF64, Covance) and rat anti-keratin8 (K8, clone TROMA-1, DSHB, University of Iowa) (Fig. 1B). Sectionswere washed three times with 0.1 M Tris buffer, stained for 20 minwith secondary antibodies, Cy3-conjugated anti-rabbit (Biolegend)and Alexa Fluor 488 goat anti-rat (Invitrogen), and washed threetimes. For tumors, the same staining procedure was followed, by usingrat anti-PECAM-1 (clone 390, Biolegend) as primary antibody andCy3-conjugated anti-rat (Invitrogen) as secondary antibody. Sectionswere counterstained for 2 min with 1 μg/ml DAPI (Biolegend), washedthree times and mounted with a coverslip (Menzel-Gläser) in Mowiolmounting medium (Calbiochem). Sections were air-dried for 1 h atroom temperature in the dark and either immediately imaged or alter-natively stored at 4 °C for few days.

Paraffin-embedded lymph node sections were processed using theautomated system Ventana Discovery. Slides were deparaffinizedusing EZprep solution. Epitope retrieval and saturation were accom-plished on the automated stainerwith CC1 solution. Endogenous perox-idases were inhibited with H2O2. Anti-CD45R/B220 primary antibody(clone RA3-6B2, BD Pharmingen) was manually deposited on slidesand incubated for 1 h. Horseradish peroxidase conjugated anti-rat sec-ondary antibody (Omnimap anti-Rat HRP) was incubated for 16 min.Slides were developed using the Chromomap DAB detection kit. Slideswere then counterstained with hematoxylin II for 16 min and post-counterstainedwith BluingReagent for 4min. Slideswere automaticallydehydrated in successive baths of absolute ethanol and histolemonusing the Leica Autostainer XL. Glass coverslips were mounted withthe Leica CV5030 module. Antibody references and dilutions are listedin Supplementary Table 1.

2.5. Automatic scanning

For the thymus and tumors, before scanning, glass slideswere cleanedwith 90% ethanol and optical cleaning tissue (VWR) to remove potentialdust or traces of mountingmedium. All specimen sections were carefullyencircled by a thin and continuous black line using a permanent marker,to entirely delineate the area to be automatically scanned. A typical one-millimeterwidthmarginwasmaintained around each specimen to avoidmissing any part of the tissue during scanning. TheMiraxmidi slide scan-ner (Zeiss) was set in the epifluorescence configuration, with properfilter-sets (Zeiss, listed in Supplementary Table 2). The microscope wasinitialized at 20× magnification using a representative slide placed into

the scanner. The protocol used to scan all sections labeled with thesame set of antibodieswas created inmanualmode using theMirax Con-trol software. Adequate emission band, detection intensity, gain andthreshold were set for each channel. Autofocus was performed on thestrongest signal, usually originating from DAPI. Scanning the trial slideallowed verification of the relevance of the settings. If required, detectionparameters were adjusted until reaching an optimal result before run-ning the protocol in automatic mode for all slides (Fig. 1C).

Lymph node sections were imaged using an automated HamamatsuNanoZoomer 2.0 HTdigital slide scanner. Slides were loaded in the scan-ner and the NDP.scan software was started. Before acquisition, slideswere entirely scanned at low magnification; lymph nodes sections wereautomatically detected and autofocus was performed. If needed, zonesof interest or supplemental focus points on the sample could be definedmanually. Zones of interest were then entirely acquired at 20× magnifi-cation, allowing a high-resolution acquisition. After scanning, high reso-lution digital images were obtained and stored in the “ndpi” format.Images were opened in Calopix Workstation (universal viewer fromTRIBVN), allowing integral visualization of sample, navigation and zoom(from 0× to 320×) of specific zones. Of note, instead of screen capture,several ImageJ plugin, such as NDPITools and ndpi_to_ome_tiff, can alsoallow direct export from “ndpi” to “tiff”.

2.6. Image registration

An image of each section was saved in “tiff” format by screen captureusing the Pannoramic Viewer free software (3DHistech). To optimizeimage resolution, for thymic and tumoral sections, a 2× magnificationwas used, leading to a pixel size of 6.5 μm. For LN sections, a 5×magnifi-cation was used, leading to a pixel size of 0.9 μm. Sequentially numberedfile names were used, according to the original order of sections. The en-tire image sequencewas imported in ImageJ (NIH, see User guide in Sup-plementary data). If necessary, for each color channel, background wasremoved and intensities were enhanced bymanually adjusting color bal-ance. If required, sections that would have been inverted duringcryosectioningwere flipped back. The “StackReg”plugin allowed registra-tion of sections respectively to each other (Thevenaz et al., 1998)(Fig. 1D). This plugin and the associated plugin “TurboReg” weredownloaded from the EPFL website, bigwww.epfl.ch/thevenaz/stackreg.For a large number of sections, the registering process could be long,and was thus first tested on a smaller copy of the image stack, scaled at10% for instance, notably to check that the resulting images remainedwithin the frame. Manual rotation or translation may also be helpful forcritical steps between highly mismatched images, as may be identifiedby this preview registration. The resulting image stack was first croppedat nearly the minimal possible size, to reduce further computation time,and then saved as a multi-tiff file.

2.7. 3D rendering

In Matlab (the Mathworks), the “rendering_3D” function was startedto reconstruct thymic lobes in 3D (Figs. 1E and 2, see also User guide inSupplementary data). If required, default parameter values were modi-fied, such as pixel size, for instance. Alternatively, for advanced userswith programing skills, values may be either modified within the codeof the associated function, “FOR3D_dialog_box”, for successive use. Themain function, “rendering_3D”, may also be run from the Matlab com-mand window, with a list of parameters in input, in the cell format, forrunning sequentially several processeswithdifferent settings for instance.Example: typing “rendering_3D({‘*.tif’; ‘130’; ‘auto’; ‘auto’; ‘6.5‘; ‘30’}) willrun For3D for every tiff file in the current folder, using a value of 130(instead of automatic evaluation) for the red threshold, automatic valuesfor the green and blue thresholds, assuming 6.5 μmpixelwidth and 30 μmsection thickness. Note that only the necessary first parameters need bedefined. Type “rendering_3D(0)” to start with all default parameters,without user validation. As for anyMatlab function, the folder containing

Page 5: For3D: Full Organ Reconstruction in 3D, an Automatized ... · For3D: Full organ reconstruction in 3D, an automatized tool for deciphering the complexity of lymphoid organs Arnauld

Fig. 2. For3D user interface. When running the “rendering_3D” function, a graphical user-friendly interface appears, listing all required parameters, together with their defaultvalues:

– Using “*.tif” for filename allows analyzing all tiff files located in the current folder.– IHC staining is detected, if relevant.– For each channel, thresholds can be automatically or manually determined.– Pixel size is set by screen capture and thickness by sectioning.– Equalization, from 1 to 0, allows total, partial, or no homogenization of the staining

over all sections.– Partial transparency allows visualization of all color channels together, i.e.1/3 for each

channel, or a lower value for large structures, extensively masking the others.– For red & green channels, non-specific staining at the border of the organ may be

eroded, assuming that the entire organ can be detected using DAPI or hematoxylincounterstaining (blue channel).

35A. Sergé et al. / Journal of Immunological Methods 424 (2015) 32–42

“rendering_3D” and the associated For3D functions must be saved eitherin the local folder or, preferentially, added to the Matlab path at first use.

For immunofluorescence signals, each staining was automaticallyassigned to a specific channel of a multicolor image, usually using theRGB (Red, Green, Blue) color space for no more than 3 channels. Forbrightfield immunohistochemistry, labels needed to be identified. Thiswas efficiently performed in the HSV (Hue, Saturation, Value) colorspace by using adequate limits for each of these three criteria (Fig. 3).A supplemental function of the For3D process, “filt_hsv”, was dedicatedfor discriminating brown and blue staining. Section sizes could presentartifactual fluctuations due to unavoidable deformations during tissuesectioning. This could affect structure realignment, identification andquantification. To correct this bias, a dedicated For3D function,“smooth_slices”, computed the size of each section and homogenizesthem, smoothing size variations over the entire stack by slightly resam-pling images. Possible staining variations among sections can be partial-ly or totally smoothed by applying an equalization step. For eachchannel, thresholds are automatically determined by For3D. However,it is a good practice to verify the automatically determined values, byusing ImageJ for instance, to (i) open the filtered image sequence,which is automatically saved by “rendering_3D” with the same name

completed by the extension “_filt.tif”, (ii) convert it to a montage toallow visualizing all images at once and (iii) check the relevance of thethreshold values (Fig. 3).

In order to correct for possible variations in staining intensity amongsections, an equalization step can beperformed, to partially homogenizemean staining levels of each section, with a strength set in the user in-terface (Fig. 2); see Irla et al. (2013) for further explanations. A 3Dview of the organ was ultimately generated by For3D (Fig. 1F) andsaved for a range of orientations in a subfolder named “3D”. 3D struc-tures are displayed with partial transparency, to allow visualization oftheir inner details. Since a strong and non-specific stainingwas regular-ly observed in all channels at the capsule level, the border, as defined byeither DAPI or hematoxylin, was eroded over a few pixels, as set by theerosion parameter (Fig. 2). Optionally, a corner could be virtually re-moved, using the “remove_3D_corner” function, to allow visualizationof the inner part of the organ (Movies 1, 3 and 4). These images werenext loaded in ImageJ and saved as a video file, in the “avi” or “mov” for-mat for instance (see User guide in Supplementary data).

For3D offers a comprehensive package for 3D reconstruction, essen-tially based on Matlab scripts, which are freely adaptable.

2.8. Substructure quantification

To quantify individual volumes, the final set of images, generated by“rendering_3D” and named with the extension “_filt.tif”, was loaded inImaris (Bitplane). The proper voxel size was adjusted, according to ex-perimental settings. For the x and y calibrations, the initial pixel size(set by screen capture: 6.5 μm for thymus and tumor, 0.9 μm for LNs)and the subsequent subsampling (as performed by “rendering_3D”: 2-fold for thymus and tumor, 4-fold for LNs) led to the final lateral size:13 μm for thymus and tumor, 3.6 μm for LNs. For the z calibration, thecryostat sectioning thickness (20 μm for thymus and tumor, 5 μm forLNs), assuming that typically 10% of sections are lost or discarded (dam-aged during sectioning or unsuitably labeled for instance) led to acorrected depth of 22 μm for thymus and tumor or 5.5 μm for LNs. Thethreshold value determined by “rendering_3D” allowed defining a vol-ume in the channel of interest, corresponding for instance to the K14staining (red channel) for the thymic medulla. This step also providedadditional verification of the adequacy of the threshold value. Surfaceand volume values were exported for further analysis, which may in-clude object enumeration, surface and volume quantification, as wellas fractal analysis, which refers to a mathematical theory dedicated tothe description of highly complex topologies, as observed for the thymicmedulla (Irla et al., 2013) or for tumor vascularization (Di Ieva, 2010).

The fractal dimension of a numeric object can be estimated as theslope of the curve displaying the number of voxels that are necessaryand sufficient for covering the object of interest, as a function of voxelsize, tested over a large range of values. This computation was per-formed in Matlab, using the “boxcount” function provided by F. Moisyon Matlab File Exchange.

2.9. Downloading For3D

The source code for For3D is available as open-source software foracademic and non-profit research upon request to A.S. ([email protected]). Scripts are providedwith a user guide (see Supple-mentary data) and a representative dataset. For3D can be appliedstraightforwardly with basic knowledge for running scientific imageanalysis software (ImageJ and Matlab), only requiring validation of theparameters in the user-friendly interface (Fig. 2). However, sinceFor3D is an open-source code for academic usage, researchers familiarwith programming are also welcome to adapt the code to match theirpersonal interest, using software such as Matlab, Imaris or ImageJ.

Detail of reagents, specific equipment and timing of the overall pro-cedure are provided in Supplementary data.

Page 6: For3D: Full Organ Reconstruction in 3D, an Automatized ... · For3D: Full organ reconstruction in 3D, an automatized tool for deciphering the complexity of lymphoid organs Arnauld

B

A

Fig. 3.Manual validation of the intensity and Hue–Saturation–Value thresholds with ImageJ. (A) Thresholding of the keratin 14 IF staining of thymic sections allows delineating the me-dulla. (B) Thresholding HSV values of IHC staining of LN sections allows discriminating either B220 staining (brown, as shown here) or counterstaining (blue, not shown). Building amon-tage of all images with ImageJ permits visualizing all sections at the same time. Thresholded areas are shown in red, and the ImageJ interface is shown next to each montage.

36 A. Sergé et al. / Journal of Immunological Methods 424 (2015) 32–42

3. Results

For3D performs reconstruction of different structures of interest,such as illustrated here for (i) the thymic corticomedullary topology,(ii) mesenteric lymph node organization and (iii) tumor vascularizationof an EG7 primary thymoma. Thymic lobes and tumors were labeled byIF and LNs by brightfield IHC.

3.1. Visualization and quantification of the thymic medulla topology

By following this method, the user should be able to visualize thecomplex 3D corticomedullary organization of themouse thymus. All thy-mic sections were first stained with anti-keratin 8 and anti-keratin 14,which specifically detect the cortex and the medulla, respectively(Fig. 4). Stained sections were then fully scanned using an automatedepifluorescence microscope, thus providing an entire view of allimmunolabeled sections. Running the For3D process allowed 3D recon-struction of the mouse thymus (Fig. 5, Movies 1 and 2). A successful

experiment with a wild-type mouse thymic lobe results in the visualiza-tion of the highly complexmedullary organization, which is composed ofa central compartment surrounded by unconnected (Fig. 5A and B,arrow) or connected (Fig. 5A and B, arrowhead) islets. For3D allowsquantification of individual medullary islet numbers as well as their re-spective volumes (Fig. 5C). A wild-type thymic lobe from an adultmouse of 6 to 8 weeks typically contains approximately 200 medullae.By analyzing several thymic lobes, we ended up with a total volume ofthe medullary compartment of 2.5 ± 1.2 mm3 (n = 3), composed of172 ± 32 individual medullary islets. Volumes are logarithmically dis-tributed, with a mean of 0.015 ± 0.017 mm3, whereas the volume ofthe large centralmedulla is on average 2±1.5mm3. Itmay beworthnot-ing that when examining sections, the medulla accounts on average for18% of the total thymic surface, while when considering the volume,due only to geometric considerations, the fraction occupied by the me-dulla is only 7.4%of the total thymic lobe. This further strengthens the im-portance of considering entire volumes, especially when dealing withcomplex geometries. While the cortex exhibited a regular geometry, we

Page 7: For3D: Full Organ Reconstruction in 3D, an Automatized ... · For3D: Full organ reconstruction in 3D, an automatized tool for deciphering the complexity of lymphoid organs Arnauld

Keratin 14 (medulla)A Keratin 8 (cortex)B DAPI (sub-capsule)C

E

M

M C

SC

MergeD

F

MC

Fig. 4.Visualization in2Dof the corticomedullary organization of themouse thymus. Thymic sections fromwild-typemice are stainedwith anti-keratin 14 antibody revealedby anti-rabbitCy3 (red) and anti-keratin 8 antibody revealed by anti-rat Alexa 488 (green), that specifically detect the medulla (A) and the cortex (B), respectively. Thymic sections are further coun-terstained with DAPI (blue) to visualize the entire thymus, and more particularly, the subcapsular region, highly dense in nuclei (C). This staining allows unambiguous identification ofeach compartment (D). Scale bar: 1 mm. (E) Zoom of the area depicted in (D). C: cortex, M: medulla, SC: subcapsule. The panel (F) shows all serial immunolabeled sections used to re-construct an entire thymic lobe by For3D.

37A. Sergé et al. / Journal of Immunological Methods 424 (2015) 32–42

obtained a fractal dimension of 2.25±0.2 (n=3) for themedulla, a frac-tional value laying between 2 and 3, emphasizing the complexity of thistopology.

3.2. Visualization and quantification of lymph node organization

In addition to IF staining, For3D can be applied to the reconstructionof organs stained by brightfield IHC. We extended the usefulness ofFor3D to other lymphoid organs, namely mouse mesenteric LNs. Paraf-fin LN sections were stained by IHC using an antibody directed againstB220, which specifically detects B cells, mainly localized at the periph-ery of the LN, in the cortex and B-cell follicles (von Andrian andMempel, 2003). Counterstaining with hematoxylin allows visualizationof the central region of the tissue, containing mainly T cells, in theparacortex andmedulla (Fig. 6). The For3D process provides a 3D recon-struction of the entire LN (Fig. 6D and Movie 3), allowing visualization

of the topology of B andT-cell zones depicted in brown and blue, respec-tively. Of note, B-follicles were specifically detected using a secondthreshold, higher than the one used for detecting the entire cortex. Ap-proximately 15 B-cell follicles are detectable as dense structures withinthe cortex (Fig. 6B and C, arrowheads). Noteworthy, taking into accountthe excellent specificity of B220 staining, sparse B cells are clearly distin-guishable within the T-cell zone (Fig. 6B and C, arrow). Thus, For3D canbe used to visualize the highly structured 3D organization of LNs.

3.3. Visualization and quantification of tumor vascularization in 3D

As demonstrated here, For3D can be applied beyond lymphoid or-gans to reconstruct mouse primary lymphoid tumors. The EG7 murinethymoma cell line was subcutaneously injected into a wild-typeC57BL/6 mouse. Two weeks later, the tumor was sectioned and recon-structed in 3D. Labeling sections for PECAM-1 provides a relevant

Page 8: For3D: Full Organ Reconstruction in 3D, an Automatized ... · For3D: Full organ reconstruction in 3D, an automatized tool for deciphering the complexity of lymphoid organs Arnauld

A C

Med

ulla

ry is

let v

olum

e (m

m3 )

Medullary isletsCentral medulla

B

Fig. 5. 3D reconstruction of a wild-type mouse thymic lobe stained by IF. (A) Thymic lobe (see Fig. 4) reconstructed by For3D. The cortical and medullary compartments are depicted ingreen and red, respectively. Axes are graduated inmillimeters. Arrowhead and arrow indicatemedullary islets connectedornot connected to themainmedulla, respectively. (B) Individualmedullary islets can bevisualizedwith Imaris, color-encoded according to their volume: from red (smallmedullae, arrow) to yellow (large centralmedulla, arrowhead), as indicatedby thecolor bar. Scale bar: 1 mm. (C) Logarithmic distribution of individual medullary islet volumes. Arrowhead and arrow indicate the volumes of the two regions respectively denoted in A.

38 A. Sergé et al. / Journal of Immunological Methods 424 (2015) 32–42

marker for tumoral vascularization (Piali et al., 1995) (Fig. 7 and Movie4). Blood vessels are observed throughout the tumor, with a gradual en-richment at the periphery. Interestingly, the gradient of vascularizationvs. tumor depth is better visualized by computing in 3D, hence over allvoxels, leading to amuch smoother curve than that obtained by compu-tation in 1D or 2D, as classically performed.

Collectively, these data illustrate that For3D can be applied to lym-phoid organs, as well as solid tumors. In addition, staining regions orcells of interest can be performed by either IF or brightfield IHC,allowing flexibility in terms of staining approaches that can be selectedby the user.

4. Discussion

We have developed a user-friendly automated method forreconstructing the 3D architecture of the mouse thymus from fluores-cence imaging of immunolabeled tissue sectionswithmultiple antibod-ies. Furthermore, we have demonstrated that For3D can be applied toother lymphoid organs such as LNs and can also be extended to otherapplications such as primary tumors. The main advantage of this ap-proach is to allow visualization and quantification of specific structuresinside an entire tissue. Of note, For3D provides good resolution of tissue

structures, because this approach relies on sectioning, and thus struc-tures located deep within the tissue are accurately revealed. For3D al-lows the 3D rotation of the reconstructed tissue, thus permittingvisualization of the tissue from different angles. The user can thereforedetermine how substructures are organized respectively to eachother, addressing for instance whether some specific structures arejoined together, and further specify and quantify their specific topology.In contrast, defining whether structures are connected to each other orform individual units cannot be readily determined by examining 2Dsections alone. As exemplified here for the mouse thymus, rotatingthe reconstructed organ demonstrates that neighboring medullary is-lets may be either connected in a continuous, labyrinth-like structure,or isolated as stand-alone units (Fig. 5A and Movie 1). This is in excel-lent accordancewith previous studies of the thymic structure, by others(Anderson et al., 2000) and us (Irla et al., 2013). Furthermore, For3D al-lows quantification of structure numbers and their respective volumes,as well as their spatial distribution. These features are perfectly exem-plified here for the thymic medulla, B-cell follicles and tumor vascular-ization (Figs. 5A, B, 6D, 7E and Movies). For example, we found around15 B-cell follicles (Fig. 6D), in agreement with others (Kumar et al.,2010) and with our previous LN reconstruction based on IF detection(Irla et al., 2013). Furthermore, the broad distribution of thymic

Page 9: For3D: Full Organ Reconstruction in 3D, an Automatized ... · For3D: Full organ reconstruction in 3D, an automatized tool for deciphering the complexity of lymphoid organs Arnauld

0.0

0.4

0.8

1.2

Vol

ume

(mm

3 )

Cortex

Follicules

Paracortex

A

D E

CB

P

C

F

F

F

F

PC

P

CF

F

Fig. 6. 3D reconstruction of a wild-type mouse mesenteric lymph node stained by brightfield IHC. (A) Entire collection of mesenteric LN sections labeled with an anti-B220 antibody andcounterstained with hematoxylin, to identify cortex/follicles (diaminobenzidine, brown) and paracortex/medulla (blue), respectively. (B) Representative image of a section located in themiddle depth of the LN. Arrowheads depict B-cell follicles and the arrow indicates isolated B cells within the paracortex. Scale bar: 1 mm. (C) Zoom of the area depicted in (B). (D) For3Dreconstruction of the LN. The cortex, follicles and paracortex aredepicted in green, red and blue, respectively. A corner is removed to allowvisualizing the internal organization of the organ.Axes are graduated in millimeters with 0.1 mm grid spacing. C: cortex, F: follicles, P: paracortex. (E) Volume quantification of each zone.

39A. Sergé et al. / Journal of Immunological Methods 424 (2015) 32–42

medullary islet volumes (Fig. 5C) directly reflects their cellularity. Thistopology and size distribution is in agreement with the notion thatmedullary compartments emerge initially from clonal expansion of sin-gle medullary thymic epithelial cell precursors (Rodewald et al., 2001).These clonal isletsmaymerge together over time, leading notably to theemergence of the large central compartment.

For3D can be highly relevant for deciphering the 3D organization ofvarious complex tissues under healthy or pathological conditions. For in-stance, the vascularization of a tumor, also exhibiting a fractal pattern, isknown to be a critical step in cancer evolution (Carmeliet and Jain, 2011).This feature can assist diagnosis bymonitoring the renormalization of thevascularization network. Vessel localization indeed classically evolvesfrom an initial chaotic, highly fractal pattern, toward a more regular andefficient network, becoming more similar to physiological organ vascu-larization. A 3D approach can provide a very accurate characterization

of the sample, as exemplified for the vascularization gradient inside atumor (Fig. 7F). Besides, this method presents critical steps that need tobe kept in mind.

4.1. Sectioning

The main drawback of For3D relates to organ sectioning, which isprone to generate artifactual distortions that may perturb rebuildingthe 3D volume from adjacent sections. However, this is handled inFor3D by adjusting the size of each image to smooth the axial profile.The For3D method is, in principle, readily applicable to any organ or tis-sue. The major issue, regarding the organs and species of interest, is ulti-mately the physical size. Indeed, to preserve a decent resolution (set bysection thickness), an organ too large (typically more than 1-cm thick-ness) may lead to a prohibitively large number of sections, hence

Page 10: For3D: Full Organ Reconstruction in 3D, an Automatized ... · For3D: Full organ reconstruction in 3D, an automatized tool for deciphering the complexity of lymphoid organs Arnauld

E

A

PECAM-1DAPI

B

C D

cd

1D gradient 2D gradient 3D gradient

Vas

cula

rized

frac

tion

Distance from tumor border (mm)

F0.3

0.2

0.1

00.5 1.510

0.3

0.2

0.1

00.5 1.510

0.3

0.2

0.1

00.5 1.510

Fig. 7. For3Dofmouse tumor vascularization stainedby IF. (A) Entire collection of EG7 thymoma sections labeledwith an anti-PECAM-1 antibody and counterstainedwithDAPI, to identifyvessels (red) and nuclei (blue), respectively. (B) Representative image of a section located in themiddle of the tumor. Scale bar: 1mm. (C, D) Zoomof the twoareas depicted in (B) near theborder and at the center of the tumor, respectively. (E) For3D reconstruction of the tumor. A corner is removed to allow visualization of the inner part of the tumor, revealing reducedvascularization. Axes are graduated in millimeters. (F) Radial distribution of the tumor vascularization: the gradient of vascularization staining was computed in 1D, 2D and 3D over allpixels/voxels equidistant from the tumor border, as defined by DAPI counterstaining. The ratio of pixels/voxels stained for PECAM-1, as defined by For3D threshold, is gradually computedfor every pixel/voxel equidistant from the tumor border. The gradient is computed using either a line scan (1D, with lateral average over 100 pixels) or by a loop of either surface (2D) orvolume (3D) erosion.

40 A. Sergé et al. / Journal of Immunological Methods 424 (2015) 32–42

increasing time and quantities of reagents. This may be avoided byaccepting a reduced number of sections, which may be achieved simplyby considering only one half of the organ, taking advantage of a suffi-ciently acceptable symmetry. On the other hand, organs too small (typi-cally below 100 μm), such as fetal organs, may be difficult to section.Alternatively, they may be reconstructed in 3D by directly acquiring aconfocal or two-photon z-stack, hence using optical instead of physicalsectioning. Furthermore, the method, being substantially time-consuming (from 7 to 34 h per organ, see Supplementary data for tim-ings), is not readily suitable for high-throughput phenotyping, but israther designed for dedicated investigations.

4.2. Immunostaining

To determine the 3D organization of specific structures, thechoice of antibodies against specific epitopes is extremely important.

To ensure substantial signals, we advise using antibodies directedagainst markers that are highly expressed in the target structure. Inall cases, and in particular when the target epitopes are weaklyexpressed, the user should opt for bright and stable dyes to efficient-ly reveal low-expressionmarkers. For IF, the user has also to considerthe endogenous autofluorescence of the target tissue. In particular,because the autofluorescence emission spectrum of biologicaltissues is often maximal at approximately 500 nm, the correspond-ing “green” channel should be used for the detection of a strongsignal, associated to an abundant epitope. Bright and photostablefluorophores, such as Alexa Fluor and cyanine, constitute goodchoices. Emission wavelengths and available filter-sets (typicallyblue, green, red and infrared) should be separated well enough inorder to limit bleed-through among channels. Finally, the use of adedicated robot (such as Ventana Discovery XT) is recommendedto improve the homogeneity of the staining.

Page 11: For3D: Full Organ Reconstruction in 3D, an Automatized ... · For3D: Full organ reconstruction in 3D, an automatized tool for deciphering the complexity of lymphoid organs Arnauld

41A. Sergé et al. / Journal of Immunological Methods 424 (2015) 32–42

4.3. Thresholding

During the analysis process, the main critical step is to set a thresh-old that accurately delimits the structure of interest. From our experi-ence, automatically setting a threshold between signal and noise is notsystematically possible, and the automatically computed value onlyprovides an initial guess for further manual adjustment. In ImageJ,converting stacks of images from each color channel to a montage al-lows viewing all images at once, which is very convenient for adjustingthe threshold of each channel (Fig. 3). Noteworthy, standardizing, oreven better, automating the labeling, directly impacts on the reproduc-ibility of the staining, hence on the ability to automatically compute arelevant threshold.

4.4. Comparison with other 3D methods

Alternative techniques for 3D visualization may provide interestingand complementary rendering, but are often restricted either in theirresolution or the accessible depth.

Tomography is available under several modalities, such as with X-rays (μCT), radio frequency (MRI), photons (OCT, OPT) (Chai et al.,2013) or positrons (PET) (Ritman, 2011;Norris et al., 2013). It has no in-herent size limit, but may provide lower resolution (typically severalmicrons) than microscopy with large samples. This 3D visualization ap-proach is very well developed for routine clinical practice, but is also afield undergoing intense development, notably toward higher contrastand resolution. However, appropriate staining and sample transparencymay constitute drawbacks that need to be considered.

Scanning microscopy, such as confocal or multi-photon microscopy(Denk et al., 1990; Helmchen and Denk, 2005), can be extended readilyfrom planar to volumetric acquisition, resulting in a stack of images(Yokomizo et al., 2012). Acquisition over a substantial number of voxelscan be time-consuming and is usually performed on fixed samples. Thistechnique may be extended to live imaging, but only under favorableconditions, i.e., those with sufficiently slow dynamics. Deconvolutionof the resulting image can further improve the resolution. However,the extent of the volume is limited by the penetration depth of light inthe specimen, typically restricted to one or a few hundred microns forconfocal or multi-photon microscopy, respectively. Furthermore, per-meabilization, and thus antibody accessibility for efficient staining,may also become difficult with increasing sample thickness. This can,however, be circumvented by using alternative staining approaches,such as transgenic expression of fluorescent proteins driven by cell- ortissue-specific promoters.

Light sheet fluorescence microscopy (Voie et al., 1993; Erturk et al.,2012; Gao et al., 2014; Reynaud et al., 2015) uses lateral planar illumina-tion, orthogonal to the optical axis, to excite awell-defined planewithinthe sample, which is typically imaged with an epifluorescence setup.This technique displays, in principle, no inherent depth limitation, andin practice can access a few millimeters. However, it requires achievinga high transparency of the specimen, which is obtained using tediousand time-consuming protocols. Furthermore, lateral illumination cannevertheless be absorbed or notably attenuated by artifacts within thespecimen. This unfortunately leads to characteristic shadow patterns,requiring the use of a second, complementary planar illuminationbeam, coming from the opposite direction, to correct this bias.

Both approaches, scanning and light sheet fluorescence microscopy,are purely optical techniques. Even though limited by optical penetra-tion depth, they can in principle investigate thick samples by optical, in-stead of physical sectioning. Hence, registration of section images is notrequired, and potential section-to-section deformations are avoided.Nevertheless, themajor advantage of For3D is to allow immunolabelingwithout depth limitation, thus providing a comprehensive view of anentire organ at depths of more than several millimeters. For3D canalso be applied to transgenic system by tagging important structural

elements or cell type of interest, such as fibroblastic reticular cells with-in lymphoid organs (Chai et al., 2013).

For3D can be readily applied to address the 3D organization of abroad range of organs, essentially providing the ability to stain struc-tures of interest. For3D thus targets a broad audience, being readily suit-able for the following biomedical fields: developmental biology,anatomical pathology, tumor biology, immunology and neurobiology.

Supplementary data to this article can be found online at http://dx.doi.org/10.1016/j.jim.2015.04.019.

Author contributions

A.S. analyzed the data, wrote the For3D code and the paper; A.L.B.performed the experiments; M.A.L. and B.A.I. gave material supportand conceptual advice. M.I. designed, performed experiments andwrote the paper.

Acknowledgments

This work was supported by the Swiss National Science Foundation(Ambizione PZ00P3_131945 to M.I. and 310030_153456 to B.A.I.), theOncosuisse (KFS-2914-02-2012 to B.A.I.), the Jules Thorn Foundation (toM.I.), the ARC Foundation (PJA 20131200238 to A.S.), the ANR (11 BSV101902 to M.A.L.), the Marie Curie Actions (Career Integration Grants,CIG_SIGnEPI4Tol_618541 to M.I.) and institutional grants from INSERM,and Aix-Marseille University. This study was also partly supported by re-search funding from the PACA Cancéropôle to set-up the IPC/CRCMExperimental Pathology Platform (ICEP) core-facility.

We would like to thank Jeanne Guenot (Geneva University), JessicaCappaï (CRCM, Marseille), the ICEP core facility (CRCM, Marseille) andthe Bioimaging core facility (GenevaUniversity) for the technical support,Pr. Graham Anderson (Birmingham University) and Dr. Pierre-HenriGaillard (CRCM, Marseille) for the critical reading of the manuscript, aswell as Prof. Walter Reith (Geneva University) for the constructive sug-gestions during the development of the For3D approach. We acknowl-edge the American Journal Experts for editorial assistance.

References

Anderson, G., Takahama, Y., 2012. Thymic epithelial cells: working class heroes for T celldevelopment and repertoire selection. Trends Immunol. 33, 256.

Anderson, M., Anderson, S.K., Farr, A.G., 2000. Thymic vasculature: organizer of the med-ullary epithelial compartment? Int. Immunol. 12, 1105.

Anderson, G., Lane, P.J., Jenkinson, E.J., 2007. Generating intrathymic microenvironmentsto establish T-cell tolerance. Nat. Rev. Immunol. 7, 954.

Carmeliet, P., Jain, R.K., 2011. Principles and mechanisms of vessel normalization for can-cer and other angiogenic diseases. Nat. Rev. Drug Discov. 10, 417.

Chai, Q., Onder, L., Scandella, E., Gil-Cruz, C., Perez-Shibayama, C., Cupovic, J., Danuser, R.,Sparwasser, T., Luther, S.A., Thiel, V., Rulicke, T., Stein, J.V., Hehlgans, T., Ludewig, B.,2013. Maturation of lymph node fibroblastic reticular cells frommyofibroblastic pre-cursors is critical for antiviral immunity. Immunity 38, 1013.

De Bock, K., Georgiadou, M., Carmeliet, P., 2013. Role of endothelial cell metabolism invessel sprouting. Cell Metab. 18, 634.

Denk, W., Strickler, J.H., Webb, W.W., 1990. Two-photon laser scanning fluorescence mi-croscopy. Science 248, 73.

Derbinski, J., Schulte, A., Kyewski, B., Klein, L., 2001. Promiscuous gene expression inmed-ullary thymic epithelial cells mirrors the peripheral self. Nat. Immunol. 2, 1032.

Di Ieva, A., 2010. Angioarchitectural morphometrics of brain tumors: are there any poten-tial histopathological biomarkers? Microvasc. Res. 80, 522.

Erturk, A., Becker, K., Jahrling, N., Mauch, C.P., Hojer, C.D., Egen, J.G., Hellal, F., Bradke, F.,Sheng, M., Dodt, H.U., 2012. Three-dimensional imaging of solvent-cleared organsusing 3DISCO. Nat. Protoc. 7, 1983.

Gao, L., Shao, L., Chen, B.C., Betzig, E., 2014. 3D live fluorescence imaging of cellular dy-namics using Bessel beam plane illumination microscopy. Nat. Protoc. 9, 1083.

Germain, R.N., Miller, M.J., Dustin, M.L., Nussenzweig, M.C., 2006. Dynamic imaging of theimmune system: progress, pitfalls and promise. Nat. Rev. Immunol. 6, 497.

Helmchen, F., Denk, W., 2005. Deep tissue two-photon microscopy. Nat. Methods 2, 932.Irla, M., Hollander, G., Reith, W., 2010. Control of central self-tolerance induction by

autoreactive CD4+ thymocytes. Trends Immunol. 31, 71.Irla, M., Guenot, J., Sealy, G., Reith, W., Imhof, B.A., Serge, A., 2013. Three-dimensional vi-

sualization of the mouse thymus organization in health and immunodeficiency.J. Immunol. 190, 586.

Page 12: For3D: Full Organ Reconstruction in 3D, an Automatized ... · For3D: Full organ reconstruction in 3D, an automatized tool for deciphering the complexity of lymphoid organs Arnauld

42 A. Sergé et al. / Journal of Immunological Methods 424 (2015) 32–42

Kumar, V., Scandella, E., Danuser, R., Onder, L., Nitschke, M., Fukui, Y., Halin, C., Ludewig,B., Stein, J.V., 2010. Global lymphoid tissue remodeling during a viral infection is or-chestrated by a B cell-lymphotoxin-dependent pathway. Blood 115, 4725.

Le Borgne, M., Ladi, E., Dzhagalov, I., Herzmark, P., Liao, Y.F., Chakraborty, A.K., Robey, E.A.,2009. The impact of negative selection on thymocyte migration in the medulla. Nat.Immunol. 10, 823.

Niesner, R.A., Hauser, A.E., 2011. Recent advances in dynamic intravital multi-photon mi-croscopy. Cytometry A 79, 789.

Norris, F.C., Wong, M.D., Greene, N.D., Scambler, P.J., Weaver, T., Weninger, W.J., Mohun,T.J., Henkelman, R.M., Lythgoe, M.F., 2013. A coming of age: advanced imaging tech-nologies for characterising the developing mouse. Trends Genet. 29, 700.

Palmer, E., 2003. Negative selection—clearing out the bad apples from the T-cell reper-toire. Nat. Rev. Immunol. 3, 383.

Pearse, G., 2006. Normal structure, function and histology of the thymus. Toxicol. Pathol.34, 504.

Piali, L., Hammel, P., Uherek, C., Bachmann, F., Gisler, R.H., Dunon, D., Imhof, B.A., 1995.CD31/PECAM-1 is a ligand for alpha v beta 3 integrin involved in adhesion of leuko-cytes to endothelium. J. Cell Biol. 130, 451.

Reynaud, E.G., Peychl, J., Huisken, J., Tomancak, P., 2015. Guide to light-sheet microscopyfor adventurous biologists. Nat. Methods 12, 30.

Ritman, E.L., 2011. Current status of developments and applications of micro-CT. Annu.Rev. Biomed. Eng. 13, 531.

Rodewald, H.R., Paul, S., Haller, C., Bluethmann, H., Blum, C., 2001. Thymus medullaconsisting of epithelial islets each derived from a single progenitor. Nature 414, 763.

Takahama, Y., 2006. Journey through the thymus: stromal guides for T-cell developmentand selection. Nat. Rev. Immunol. 6, 127.

Thevenaz, P., Ruttimann, U.E., Unser, M., 1998. A pyramid approach to subpixel registra-tion based on intensity. IEEE Trans. Image Process. 7, 27.

Voie, A.H., Burns, D.H., Spelman, F.A., 1993. Orthogonal-plane fluorescence optical sec-tioning: three-dimensional imaging of macroscopic biological specimens. J. Microsc.170, 229.

von Andrian, U.H., Mempel, T.R., 2003. Homing and cellular traffic in lymph nodes. Nat.Rev. Immunol. 3, 867.

Welti, J., Loges, S., Dimmeler, S., Carmeliet, P., 2013. Recent molecular discoveries in an-giogenesis and antiangiogenic therapies in cancer. J. Clin. Invest. 123, 3190.

Willard-Mack, C.L., 2006. Normal structure, function, and histology of lymph nodes.Toxicol. Pathol. 34, 409.

Yokomizo, T., Yamada-Inagawa, T., Yzaguirre, A.D., Chen, M.J., Speck, N.A., Dzierzak, E.,2012. Whole-mount three-dimensional imaging of internally localized immuno-stained cells within mouse embryos. Nat. Protoc. 7, 421.