assembling models of embryo development: image …assembling models of embryo development: image...

Assembling Models of Embryo Development:Image Analysis and the Construction of DigitalAtlases

Carlos Castro-Gonzalez, Marıa J. Ledesma-Carbayo,Nadine Peyrieras, and Andres Santos*

Digital atlases of animal development provide a quantitative descriptionof morphogenesis, opening the path toward processes modeling. Proto-typic atlases offer a data integration framework where to gather infor-mation from cohorts of individuals with phenotypic variability. Relevantinformation for further theoretical reconstruction includes measure-ments in time and space for cell behaviors and gene expression. The lat-ter as well as data integration in a prototypic model, rely on image proc-essing strategies. Developing the tools to integrate and analyze biologi-cal multidimensional data are highly relevant for assessing chemicaltoxicity or performing drugs preclinical testing. This article surveyssome of the most prominent efforts to assemble these prototypes, cate-gorizes them according to salient criteria and discusses the key ques-tions in the field and the future challenges toward the reconstruction ofmultiscale dynamics in model organisms. Birth Defects Research(Part C) 96:109–120, 2012. VC 2012 Wiley Periodicals, Inc.

Key words: digital atlases; embryo development; image processing; intoto microscopy imaging

INTRODUCTIONStudying how the known genomesrelate to the spatiotemporal behav-ior of cell dynamics, tissue pattern-ing, and embryo morphogenesis isone of the key questions for devel-opmental biology in the postgenomicera. Despite completion of the ge-nome sequence of many organisms(McPherson et al., 2001), we are stillfar from understanding, modeling,and predicting how organisms de-velop from one single cell into anorganized, multicellular individual.However, comprehensive under-

standing of biological mechanismsis a fundamental issue for efficient

preclinical testing of potential newdrugs (Goldsmith, 2004; Sipeset al., 2011). Potential applicationsinclude treatment of heart diseases(Milan et al., 2003; Barros et al.,2008), leukemia (North et al., 2007),bone disorders (Paul et al., 2008),cancer (Amatruda et al., 2002; Luet al., 2011), schizophrenia, Parkin-son’s, Alzheimer’s, and other demen-tia (Martone et al., 2008).Many fundamental challenges

pave the way toward the long termgoal of reconstructing living systemsmultiscale dynamics. The quantita-tive assessment of the temporal andspatial gene expression distribution

in multicellular organisms is requiredfor building and modeling gene reg-ulatory networks underlying mor-phogenesis (Davidson and Erwin,2006; Li and Davidson, 2009).Recent advances in labeling tech-

niques (Vonesch et al., 2006; Choiet al., 2010) and microscopic imag-ing (Megason and Fraser, 2007)have steered this field from a static,‘‘omics’’-like approach (Walter et al.,2002) toward image-based strat-egies providing spatial and temporalquantitative information (Fer-nandez-Gonzalez et al., 2006; Gor-finkiel et al., 2011). Hence, the cur-rent trend toward automatic, high-content, high-throughput screeningbrings new bottlenecks in the do-main of image analysis (Baker,2010; Truong and Supatto, 2011):The unprecedented rise in complex-ity and size of data have favored theblossoming of a new discipline, bio-image informatics (Peng, 2008), orthe science of organizing distributedand heterogeneous biological imagedata into typed data and categorizedquantitative information.In particular, this review deals

with the recent strategies developedto achieve the reconstruction of dig-ital anatomy and gene expressionatlases for different model organ-isms (Table 1). The reconstruction

REVIE

W

VC 2012 Wiley Periodicals, Inc.

Birth Defects Research (Part C) 96:109–120 (2012)

Carlos Castro-Gonzalez, Marıa J. Ledesma-Carbayo, and Andres Santos are from Biomedical Image Technologies, ETSIT, Uni-versidad Politecnica de Madrid, Spain and Biomedical Research Center in Bioengineering, Biomaterials and Nanomedicine (CIBER-BBN), Spain.Nadine Peyrieras is from N&D, Institut de Neurobiologie Alfred Fessard, Gif-sur-Yvette, France

Supported by grants from the Spanish Ministry of Science and Innovation TEC2010-21619-C04-03, TEC2011-28972-C02-02, CDTI-CENIT (AMIT), INNPACTO (PRECISION), Comunidad de Madrid (ARTEMIS), French Programs ANR and ARC, European Regional De-velopment Funds (FEDER), FP6 New Emerging Science and Technology EC Program and FP7 Health program.

*Correspondence to: Andres Santos or Nadine Peyrieras, ETSIT-UPM, Biomedical Images Technologies (BIT), Madrid, Spain.E-mail: [email protected]

View this article online at (wileyonlinelibrary.com). DOI: 10.1002/bdrc.21012

TABLE

1.Overv

iew

ofRecentStrate

gie

sfo

rReconstructing

Dig

italAnato

myand

GeneExpre

ssio

nAtlasesfo

rAnim

alOrg

anism

s

Refere

nce

Anim

al

model

Imaging

modality

Spatialsc

opeand

reso

lution

Developmentalperiod,

ageand#

ofsteps

Data

type:

#ofsp

ecim

ens,

template

type,

#ofgen

esandgen

eproduct

quantifica

tion

Matchingtypeand

transform

ation

Murrayetal.

(2008)

C.elegans

Invivo

micro

scopy

Whole

org

anism

Cellu

lar

Early

4–350

cells

5hrc

20

Individual

4Yes

Object-

base

dNonrigid

Longetal.

(2009)

C.elegans

Micro

scopy

Whole

org

anism

Cellu

lar

Early

L11

15

Individual

0No

Object-

base

dAffine

Liuetal.

(2009)

C.elegans

Micro

scopy

Whole

org

anism

Cellu

lar

Early

L11

324

Synth

etic

93

Yes

Object-

base

dNonrigid

Tomeretal.

(2010)

Platynere

isMicro

scopy

Bra

inCellu

lar

Late

48hr

1171

Avera

ge

8No

Intensity-

base

dAffine1

nonrigid

Fowlkesetal.

(2008)

Dro

sophila

Micro

scopy

Whole

org

anism

Cellu

lar

Early

0–100%

61822

Avera

ge

95

Yes

Object-

base

dNonrigid

Friseetal.

(2010)a

Dro

sophila

Micro

scopy

Whole

org

anism

Multicellu

lar

Early

5%

12693

Synth

etic

1881

Yes

Object-

base

dNonrigid

Pengetal.

(2011)

Dro

sophila

Micro

scopy

Bra

inTissu

lar

Adulthood

Adult

12945

Avera

ge

470

No

Object-

base

dAffine1

nonrigid

Potikanond

etal.

(2011)

Zebra

fish

Micro

scopy

Whole

org

anism

Tissu

lar

Late

24–72hr

4�7

5In

dividual

�20

No

Semantic

-

Rittsch

er

etal.

(2011)

Zebra

fish

Micro

scopy

Whole

org

anism

Org

anular

Late

120hr

111

Avera

ge

0No

Object-

base

dRigid

1nonrigid

Ullm

anetal.

(2010)

Zebra

fish

MRI

Bra

inTissu

lar

Late

4month

s1

1b

Individual

0No

None

None

Castro

etal.

(2009)

Zebra

fish

Micro

scopy

Whole

org

anism

Cellu

lar

Early

6hr

16

Individual

5Yes

Intensity-

base

dRigid

Ruffinsetal.

(2007)

Quail

MRI

Whole

org

anism

Org

anular

Late

5–10days

66b

Individual

0No

None

None

Fish

eretal.

(2011)

Chicken

OPT

Wingbud

Tissu

lar

Late

3–5days

345

Individual

5No

Object-

base

dNonrigid

Baldock

etal.

(2003)a

Mouse

OPT

Whole

org

anism

Org

anular

Late

2–17days

22

23,484

Individual

14927

No

Object-

base

dNonrigid

Carsonetal.

(2005)

Mouse

Micro

scopy

Bra

inCellu

lar

Late

7days

1200

Synth

etic

200

Yes

Object-

base

dAffine1

nonrigid

Kovace

vic

etal.

(2005)

Mouse

MRI

Bra

inTissu

lar

Late

56days

19

Avera

ge

0No

Intensity-

base

dAffine1

nonrigid

Maetal.

(2005)

Mouse

MRI

Bra

inTissu

lar

Late

56days

110

Pro

babilistic

0No

Intensity-

base

dAffine1

nonrigid

Johnso

netal.

(2010)

Mouse

MRI

Bra

inTissu

lar

Late

56days

114

Pro

babilistic

0No

Intensity-

base

dAffine

Lein

etal.

(2007)a

Mouse

Micro

scopy1

MRI

Bra

inCellu

lar

Late

56days

1�2

0,000

Avera

ge

�20000

Yes

Intensity-

base

dRigid

1nonrigid

Woodsetal.

(1999)

Human

MRI

Bra

inOrg

anular

Adulthood

Adult

122

Avera

ge

0No

Intensity-

base

dNonrigid

Rexetal.

(2003)

Human

MRI

Bra

inTissu

lar

Adulthood

Young

adult

1452

Avera

ge

0No

Intensity-

base

dAffine1

nonrigid

Smithetal.

(2004)

Human

MRI

Bra

inTissu

lar

Adulthood

Adult

158

Avera

ge

0No

Intensity-

base

dAffine

Kerw

inetal.

(2010)

Human

OPT

Bra

inTissu

lar

Late

47–55

days

25

Individual

3No

Object-

base

dNonrigid

hr5

hours,%

5percentageofmembra

neinvagination.

a2D

geneexpre

ssions.

bNomatchingpro

cedure

:Each

specimenwasdirectly

employedasth

etemplate

ofitsco

rresp

ondingdevelopmentalstage.

c3D+tliv

eim

aging.

of a digital atlas requires a series ofimage processing steps (‘‘Proposedimage processing pipeline’’ section)to map a cohort of individuals onto acommon reference space. Theseoperations allow the investigator tocombine unrelated data and to pro-vide a single representation to visu-alize, mine, correlate, and interpretinformation at different scales.The result is the assembly of a

digital prototypic model of a ‘‘stand-ard’’ individual which constitutesthe essential scaffold where tomake the accurate, repeatable,consistent, and quantitative meas-ures required for comparative stud-ies (Oates et al., 2009). Atlases canbe compared to geographic infor-mation systems (GIS): ‘‘spatialdatabases to which diverse data,primarily but not restricted to imag-ing data, can be registered andqueried’’ (Martone et al., 2008). Forexample, atlases are used to iden-tify and categorize anatomic andgenetic differences between cohortsof individuals, such as different mu-tant strains (Warga and Kane,2003) and constitute an essentialtool that allows relating genotypesand phenotypes. This review isorganized as follows: ‘‘Classificationof digital atlases, Animal models,Imaging modalities, Spatial scopesand resolutions, Developmentalstages, Data types, and Matchingprocedures’’ sections discuss recenttrends in the field and propose a clas-sification of anatomy and geneexpression atlases for model organ-isms and human based on variouscriteria. ‘‘Proposed image processingpipeline and Visualization and valida-tion’’ sections describe a genericimage processing framework toreconstruct, validate, visualize, andinteract with a digital model ofembryo development. ‘‘Biologicalinsights’’ section surveys some of thebiological insights that can be derivedfrom such atlases. Finally, ‘‘Perspec-tives’’ section deals with the discus-sion and perspectives on the subject.

CLASSIFICATION OF

DIGITAL ATLASES

We propose a classification of digitalanatomy and gene expressionatlases for animal organisms based

on the following ontology (Table 1):(1) Animal models, (2) imagingmodalities, (3) spatial scopes andresolutions, (4) developmentalstages, (5) data types, and (6)matching procedures. In the follow-ing sections, we will discuss theseseparate criteria in more detail.The construction of anatomy

atlases and the development ofappropriate computation strategiesis a major issue in the medical field(Park et al., 2003; Aljabar et al.,2009; Fonseca et al., 2011). Theconstruction of human brain atlasesreceived special attention and manyalgorithmic reconstruction and visu-alization methods and tools comefrom this field (Mazziotta et al.,2001; Toga et al., 2006). The studyof model organisms allowed explor-ing fine spatial and temporal scalesand aimed at gathering an increas-ing amount of information includinggene expression data. We focushere on many model organisms(‘‘Animal models’’ section) imagedwith three different image modal-ities (‘‘Imaging modalities’’ section).We distinguish between atlases lim-ited to the brain and atlases encom-passing the whole organism (‘‘Spa-tial scopes and resolutions’’ section),either at the adult stage or through-out embryonic stages (‘‘Develop-mental stages’’ section) and focus-ing either on anatomical structuresor gathering gene expression data(‘‘Data types’’ section). We also con-sider different strategies to matchindividuals into the atlas model(‘‘Matching procedures’’ section).

ANIMAL MODELS

Model organisms are chosen fortheir small size, good properties interms of phylogenetic position (Fig.1), transparency, and/or relevancefor studies related to human health.The nematode Coenorhabditis ele-

gans, having the most ancient evolu-tionary emergence among the con-sidered animal models, has a largelyinvariant cell lineage and stereotypeddevelopment which greatly facilitatescomparisons between different indi-viduals (Murray et al., 2008; Liuet al., 2009; Long et al., 2009).The worm Platynereis kept sev-

eral ancestral traits (Tomer et al.,

2010) and proved being insightfulfor comparative studies.The fruit fly Drosophila mela-

nogaster has been extensivelystudied in the field of genetics anddevelopmental biology (Fowlkeset al., 2008; Frise et al., 2010;Peng et al., 2011). Sixty percentof so called genetic diseases inhumans have their counterpart inthe Drosophila genome.The zebrafish (Danio rerio) has

more recently emerged as a modelfor developmental biology researchbecause of its amenability togenetic investigations and thetransparency of its tissues. In addi-tion, its closer phylogenetic positionto humanmakes it a valuable modelfor toxicology and pharmacologystudies (Hill et al., 2005; Yanget al., 2009). Anatomy and geneexpression atlases for the zebrafishbrain or whole organism at differentdevelopmental stages are under-way (Castro et al., 2009; Ullmannet al., 2010; Potikanond and Ver-beek, 2011; Rittscher et al., 2011).Quail (Ruffins et al., 2007) and

chicken (Fisher et al., 2008; Fisheret al., 2011) are also used as ver-tebrate models and have interest-ing features for experimental em-bryology. The embryo can developoutside the egg, is quite well ame-nable to in vivo imaging (‘‘Imagingmodalities’’ section) and the con-struction of atlases with cellularresolution (see ‘‘Spatial scopesand resolutions’’ section).Mouse is the major mammalian

model organism for biomedicalinvestigations and much effort hasbeen devoted to the reconstructionof their development (MacKenzie-Graham et al., 2004; Carson et al.,2005; Kovac;evic; et al., 2005; Maet al., 2005; Lein et al., 2007; John-son et al., 2010; Richardson et al.,2010; Hawrylycz et al., 2011). Therelatively large size of the mouseembryo makes it difficult to capturethe whole specimen in a single-shot,in-toto imaging strategy (‘‘Imagingmodalities’’ section) with sufficientspatial resolution (‘‘Spatial scopesand resolutions’’ section).The same difficulty applies to

fixed human embryos (Woodset al., 1999; Rex et al., 2003;Smith et al., 2004; Kerwin et al.,

ASSEMBLING MODELS OF EMBRYO DEVELOPMENT 111


2010), where the creation of astandard cartography of humanbrains is of fundamental impor-tance in medical studies.

IMAGING MODALITIES

Three main image modalities havebeen employed in the assembly ofdigital atlases: Fluorescence mi-croscopy, magnetic resonanceimaging (MRI) and optical projec-tion tomography (OPT).Each of these modalities has dif-

ferent optical resolutions and leadto different types of atlases (Table1). The choice depends on thespecimen thickness and its opticaltransparency. For each animalmodel (‘‘Animal models’’ section)these properties vary with the ageof the specimen (‘‘Developmentalstages’’ section).Recent advances in photonic mi-

croscopy imaging (Fig. 2A) includemultiharmonic (Evanko et al.,2010) and fluorescence imaging byconfocal, multiphoton laser scan-ning microscopy (Abbott, 2009;Pardo-Martin et al., 2010) or light-sheet fluorescence microscopy

(LSFM) (Huisken and Stainier,2009; Keller et al., 2010), com-bined with newly developed fluo-rescent proteins and biological sen-sors (Chudakov et al., 2005; Giep-mans et al., 2006) and in situhybridization (ISH) techniques(Welten et al., 2006; Brend andHolley, 2009). These advancesopened new perspectives for theconstruction of high resolution ana-tomical and gene expressionatlases. Spatial resolution of hun-dreds of nanometers and temporalresolution of minutes have beenachieved for the observation ofentire organisms at different levelsof organization. However, photonicmicroscopy imaging is still limitedto small model organisms withgood optical properties.OPT (Sharpe et al., 2002) was

introduced as an alternative opticalmethod to fluorescence microscopyand overcomes the limitation of thespecimen thickness. OPT generatesdata by acquiring many views of thesame specimen at different rotationangles then assembled to create a3D volume (Fig. 2C). OPT resolutionin the range of millimeters does not

however allow working at the singlecell level.Alternatively, MRI (Jacobs et al.,

2003) does not use fluorescentstaining and has thus a broad rangeof applications (Fig. 2B). Indeed,MRI contrast does not depend on thepenetration of photons but on thevoxel-to-voxel variations in watercontent leading to diverging spinswhen submitted to magnetic fields.MRI achieves a spatial resolution ofabout tens of microns only, andalthough more and more intensemagnetic fields are used, single cellresolution is barely achieved.

SPATIAL SCOPES AND

RESOLUTIONS

Constructing a prototypic model foran organism can achieve differentscopes, from particular organs tothe whole organism, which can beresolved at either the organ, tissue,multicellular, or eventually cellularresolutions (Table 1). The recon-struction of atlases with resolutionat the cellular level (Fig. 2D) focusedon the species more phylogeneti-cally distant from human (‘‘Animal

Figure 1. Timing of evolutionary emergence and phylogenetic relationships of different model organisms. Time estimations wereextracted from Hedges et al. (2006).

112 CASTRO-GONZALEZ ET AL.


Figure 2. Examples of components involved in an atlas model construction. A: Confocal microscopy acquisition of a 24 hours postfertilization (hpf) zebrafish brain labeled by fluorescent ISH (tyrosine hydroxylase RNA probe) and DAPI staining of cell nuclei. Scalebar 100 microns. B: MRI of a quail extracted from Caltech’s ‘‘Quail Developmental Atlas.’’ Available from: http://131.215.15.121/.C: OPT of a late mouse embryo extracted from the ‘‘EMAP eMouse Atlas Project,’’ http://www.emouseatlas.org. Scale bar: 1000microns. D: Orthoslice showing the nuclei of a zebrafish early embryo where the raw gene expression from another specimen hasbeen integrated. Cells positive for the expression of the gene are highlighted in blue. Scale bar: 100 microns. E: Zebrafish tem-plates for three different developmental stages where individuals can be mapped using a reference gene pattern. F: Reconstructionof a mosaic-like atlas: Guided by a reference pattern, partial views of different individuals are mapped into a complete template. G:Left panel, coronal section of an averaged 3D template showing organ-level anatomical annotations of an adult mouse brain at agiven developmental stage. Right panel, an ISH slice warped into the atlas template through deformable models. Extracted fromthe ‘‘Allen Mouse Brain Atlas [Internet]. Seattle (WA): Allen Institute for Brain Science. 2009.’’ Available from: http://mouse.brain-map.org. Scale bar: 1300 microns.



models’’ section), and addressedrather early developmental stages(‘‘Developmental stages’’ section).Identifying every single cell positionin the whole imaged specimen(Long et al., 2009) requiresadvanced image processing meth-ods (‘‘Proposed image processingpipeline’’ section). Algorithmic strat-egies for the approximate detectionof the cell nuclei center in 3D vol-umes encompassing several thou-sands of cells have been described(Drblikova et al., 2007; Frolkovicet al., 2007; Kriva et al., 2010). Inaddition, the identification of cellcontours helps assigning RNAs orprotein expression to the cell. Voro-noi geometries have been proposedas a simple approach to determinecellular boundaries (Luengo-Orozet al., 2008). The cell shape can bet-ter be obtained by the algorithmicsegmentation of cell membranes in3D when the latter is available(Zanella et al., 2010; Mikula et al.,2011).Working at the mesoscopic scale

of the multicellular structure is lessdemanding and has already pro-vided useful information (Fisheret al., 2008; Frise et al., 2010).Annotating and segmenting the

different anatomical structures ofinterest at the tissue level isrequired to reconstruct prototypicmodels of organs. Examples of suchmethods can be found in Ma et al.(2005), Kovacevic et al. (2005),Dorr et al. (2008), Johnson et al.(2010), and Ullmann et al. (2010).Finally, large organisms with a

huge number of cells and high com-plexity in terms of organization ledto organ-level atlases that do notresolve the single cell level (Fig.2G). This strategy has been usedfor vertebrates at late developmen-tal stages (‘‘Developmental stages’’section) when the specimen’s sizeand lack of optical transparency donot allow imaging with resolution atthe single cell level (Baldock et al.,2003; Ruffins et al., 2007; Rittscheret al., 2011).

DEVELOPMENTAL STAGES

The construction of anatomical andgene expression atlases focused on

early developmental stages as wellas adulthood (Table 1). At early de-velopmental stages, the whole or-ganism is more easily amenable toin toto imaging with resolution atthe single cell level, Figure 2D, E(Fowlkes et al., 2008; Castro et al.,2009; Long et al., 2009).At later developmental stages or

in the adult, it can be more rele-vant to focus on specific organs(Fig. 2B, C) such as the brain(Woods et al., 1999; Lein et al.,2007; Peng et al., 2011) orappendages (Fisher et al., 2011).Most studies targeted a single de-

velopmental stage (Fig. 2G). How-ever, the temporal scale is essentialto the understanding of biologicalmechanisms and gathering atlaseswith the relevant kinetics is a majorissue in the field (Fig. 2E). Fisheret al. (2011) reconstructed atlasescombining fate mapping data andgene expression patterns for threeconsecutive developmental stagesof the chick wing bud. Murray et al.(2008) took advantage of thelargely invariant lineage of C. ele-gans to build the first 3D1time atlasof transgenic reporters’ expressionpatterns in C. elegans from the 4-cell stage to the 350-cell stage.

DATA TYPES

Many different specimens areassembled in the construction ofatlases models that can just carryanatomical information or multile-vel, genomewide data.Anatomical atlases providing a

scaffold with the morphological andhistological landmarks characteristicof a cohort (Rex et al., 2003; Ruffinset al., 2007; Ullmann et al., 2010)constitute a reference shape or tem-plate to integrate further informationcoming from other individuals andreflect the intrinsic multilevel of mor-phogenesis processes. Genomewideatlases (Fisher et al., 2008; Richard-son et al., 2010) integrate geneexpression patterns and multilevelinformation from various sourcesinto anatomical atlases (Fig. 2D).This approach emulates a virtualmultiplexing and overcomes therestrictions in the number of geneproducts and/or functional patterns

that can be simultaneously assessed.As a consequence, they are becom-ing a major tool for making spatio-temporal correlations between thedifferent levels of biological organiza-tion, comparing individuals, buildingprototypic models, and decipheringthe relationship between genotypesand phenotypes.Building an anatomical atlas

requires defining a common scaf-fold, frequently called template,where to gather all the informationcollected from different specimens.There are diverging criteria in theliterature about how an atlas tem-plate should be built. Several stud-ies (Ruffins et al., 2007; Castroet al., 2009; Ullmann et al., 2010)employed one single individual tomatch all the rest of the population(Fig. 2D). This individual is chosenfor its ‘‘standard’’ appearance andthe corresponding data should beof the highest quality. Alterna-tively, an iterative method hasbeen used to identify the medianindividual within a population andselect it as the template (Longet al., 2009). Other projects (Friseet al., 2010) used a synthetic tem-plate to map all the data from acohort of specimens. This templateconsists in an engineered ‘‘virtualspecimen’’ which retains the essen-tial features of a species. The use ofan average template (Fig. 2G) iswidely spread (Rex et al., 2003;Fowlkes et al., 2008; Peng et al.,2011). Ma et al. (2005) con-structed a ‘‘minimal deformationaverage template’’ as an idealizedspecimen minimizing the deforma-tion required to fit any specimen ofthe cohort. Although average tem-plates usually imply a better signal-to-noise ratio than individual speci-mens and exhibit a better definitionin very similar regions betweenspecimens, they fail to faithfullymodel fine features and regionswith a high variability, loweringtheir definition (Kovacevic et al.,2005; Dorr et al., 2008). Finally,some approaches used a probabil-istic template (Johnson et al.,2010) where specimens’ variabilityis represented by statistical confi-dence limits.The construction of prototypical

genomewide atlases implies imag-



ing gene expression patterns in 3Dwith resolution at the cellular level(Hendriks et al., 2006). Imageprocessing methods are required toachieve the automated segmenta-tion of gene expression domainsand the quantification of gene prod-ucts to allow for example thedescription of expression domainborders. A simple quantificationstrategy is based on the assumptionof a linear relationship between flu-orescence intensity and geneexpression level (Wu and Pollard,2005; Frise et al., 2010). Theobtained measurements are oftennormalized with respect to thenuclei channel fluorescence (Liuet al., 2009), which is considered tobe constant, to compensate forthickness-dependent signal detec-tion. Normalization with respect tothe background intensity (Murrayet al., 2008) is also a common strat-egy. Another possibility consists inclustering a population of cells intodiscrete levels (e.g. strong, moder-ate, weak, and none) depending onthe gene expression signal intensity(Carson et al., 2005). Although thethree different methods yielded cor-related measurements across dif-ferent individuals, the relevance ofthe obtained quantitative measure-ments to compare different speci-mens is questionable and this issueremains a challenge. Current effortsto achieve the quantitative compar-ison of gene expression levels in dif-ferent individuals include the mini-mization of variability within a pop-ulation (Fowlkes et al., 2008) andthe conversion of fluorescence sig-nal into fluorescent proteins num-ber in transgenic specimens (Damleet al., 2006).

MATCHING PROCEDURES

A matching procedure is required toimport each specimen (the source)into the template according to themaximization of a likelihood criteria.Repeating this operation is the coreof digital atlases construction. Forthe same purpose, medical imagingmakes extensive use of registrationtechniques (Maintz and Viergever,1998; Zitova and Flusser, 2003).Three main registration techniquesto build digital atlases can be distin-

guished according to the informationused to assemble the data and theminimization criteria chosen accord-ingly: Intensity-based, object-based, and semantic-based registra-tions (‘‘Intensity-based registration,Object-based registration, andSemantic-based registration’’ sub-sections). We can also distinguishthree different transformation typesbetween the source and the tem-plate space: Rigid, affine or nonrigid(‘‘Transformation categories’’ sub-section).Before the registration step, an

initialization scheme is generallyapplied to get a rough alignmentbetween source and template. Theinitialization scheme helps theregistration to reach an accuratesolution. Two common initializa-tion techniques consist of coarselyaligning anatomical landmarks(Lein et al., 2007) or the majororientation axis of an organismsuch as the anterior–posterior ordorsal–ventral axis (Blanchoudet al., 2010). Qu and Peng (2010)developed an original skeletonstandardization technique to ruleout part of the geometrical vari-ability between Drosophilaembryos. In the same line, Penget al. (2008) designed a methodto straighten C. elegans wormsinto the same canonical space.The populations of individuals to

be registered are normally composedof complete specimens imaged simi-larly. Accurately matching cohorts ofpartial specimens (Fig. 2F) is one ofthe current challenges in the field(Peng et al., 2011) and very fewstrategies addressed this case(Castro et al., 2009).

Transformation Categories

Rigid transformations are appliedwhen the mapping between thesource and template spaces con-sists of spatial translations androtations (Castro et al., 2009). Rigidregistration has the advantage ofkeeping the original raw data unal-tered, allowing faithful measure-ments and validation of the truevolumes in the final atlas represen-tation. Affine transformations (Rexet al., 2003; Smith et al., 2004)also include a scaling factor in addi-

tion to translations and rotations.Both rigid and affine transforma-tions are linear and globally appliedto all voxels.On the contrary, nonrigid trans-

formations are nonlinear andlocally warp the source image tofit into the template (Woods et al.,1998; Ng et al., 2007; Ng et al.,2009; Rittscher et al., 2010). Thistypically results in an alteration ofthe original raw data.

Intensity-Based Registration

Intensity-based registration pro-cedures align the source and tem-plate by trying to maximize a simi-larity metric (typically mutual in-formation or cross correlation)between the gray level values inthe voxels of both images.The most common approaches

(Lein et al., 2007; Tomer et al.,2010) include an initialization per-formed by a global, intensity-basedaffine or rigid registration, followedby local deformable warps (Fig. 2G).Multiresolution approaches are alsoemployed to optimize the mappingprocedure in a coarse-to-fine strat-egy (Smith et al., 2004; Kovacevicet al., 2005; Tomer et al., 2010).Finally, multimodal approachescombine information coming fromdifferent imaging modalities, merg-ing, for instance, histology and MRI(MacKenzie-Graham et al., 2004;Johnson et al., 2010). Suchapproaches provide multiple entrypoints to match different individualsand heterogeneous populations intothe same coordinate system.

Object-Based Registration

Object-based transformationsattempt to bring into alignmentequivalent sets of characteristicpoints or landmarks present in boththe source and template images(Liu et al., 2009). These transfor-mations are local and nonlinear andtypically produce an alteration ofthe data shapes and volumes. Penget al. (2011) developed an auto-matic pattern recognition system toidentify and match visual anatomicreferences with certain geometricproperties such as high local curva-ture, and Fowlkes et al. (2008) cre-



ated a method to unequivocallyidentify the cell to cell correspon-dence in C. elegans embryos.

Semantic-Based Registration

Unlike intensity-based and object-based registrations, semantic-basedregistration does not operate on thegeometrical space and is based onthe use of standardized ontologies(Ashburner et al., 2000) and webqueries (Zaslavsky et al., 2004;Potikanond and Verbeek, 2011).After following an annotation

procedure for anatomy and geneexpression data with a controlled,standard vocabulary, the mappingprocedure is reduced to just link-ing names to positions or domains(Baldock et al., 2003; Boline et al.,2008). Given the difficulty of geo-metric registration across greatlyvariable resources, this strategy isuseful to guarantee interoperabilityand can bring together data com-ing from different laboratories,resources, developmental stages,or even different species.

PROPOSED IMAGE

PROCESSING PIPELINE

The construction of atlases or digi-tal representations of anatomicand genetic features from anincreasing amount of more andmore complex data, requires so-phisticated image analysis algo-rithms (Khairy and Keller, 2011)replacing nonefficient and timeconsuming processing performedmanually or through generic imag-ing software.We describe a generalized image

processing pipeline to gatherquantitative, genomewide datafrom a cohort of individuals in aprototype with resolution at thecellular level (Fig. 3). This pipelinecan achieve a complete (x, y, z,G) model, including quantitativedata for gene or protein expres-sion level (G) in each cell position(x, y, z) in a developing model or-ganism (Castro et al., 2011).This process can involve prepro-

cessing steps, such as imageenhancement or multiviews fusionalgorithms (Rubio-Guivernauet al., 2012). Then, cell nuclei

detection and cell segmentationtechniques (‘‘Spatial scopes andresolutions’’ section) are appliedto one embryo, i, to extract cellposition, (xi, yi, zi), and volume.Next, signal quantification proce-dures (‘‘Data type’’ section) areapplied on gene expression, gi, toidentify positive cells, (xi, yi, zi,gi). Repeating this procedure forall the individuals of a cohort, 1 toN, yields measurements for rele-vant patterns, (x1, y1, z1,g1). . .(xN, yN, zN, gN), which arefinally combined through a regis-tration procedure (‘‘Matching pro-cedures’’ section). The anatomicalinformation extracted during thecell detection step and automati-cally segmented or identified land-marks guide the registration pro-cess. The final result is a single,quantitative model of the speci-men development, (x, y, z, g1. . .gN). Validation of the model andfurther analysis use a dedicated,custom-made interactive visual-ization interface (‘‘Visualizationand validation’’ section).

VISUALIZATION AND

VALIDATION

The reconstruction of digitalatlases relies on automatic algo-rithms that can handle the enor-mous amount of large 3D imagesproviding multilevel data forcohorts of individuals at differentdevelopmental stages. The lack ofgold standards in the field requiresthe manual curation and correc-tion of the results (Long et al.,2009).Several indirect validation tech-

niques have been exploited:Fowlkes et al. (2008) and Penget al. (2011) showed that thegene expression variability in theiratlas model was comparable tothat shown by individuals, imply-ing that the experimental errorsintroduced in the model could beconsidered negligible. Fisher et al.(2011) applied hierarchical clus-tering (Pearson Correlation) toreplicates coming from differentspecimens and found that theysegregated as expected. In addi-tion to these indirect validationmeasures, visual assessment is

the common validation standardfor virtually all the previouslydescribed strategies (Table 1).Consequently, many sophisticatedvisualization platforms have beendeveloped to display the multidi-mensional input data and outputresults, interactively run the previ-ously described methods onrequest while providing the neces-sary tools to correct, annotate,quantify, and mine their out-comes. These platforms representthe necessary trade-off betweenthe automated, high-throughput,fast computer algorithms and themanual, low-throughput but accu-rate human interactions.A comprehensive review of such

visualization tools can be found inWalter et al. (2010). Some relevantinstances include: FlyEx (Pisarevet al., 2009), GoFigure (Gouaillardet al., 2007), PointCloudExplore(Weber et al., 2009; Rubel et al.,2010), Mov-IT (Olivier et al., 2010),BrainExplorer (Lau et al., 2008),BrainGazer (Bruckner et al., 2009),CellProfiler (Jones et al., 2008), andV3D (Peng et al., 2010).

BIOLOGICAL INSIGHTS

The application of image process-ing tools (see ‘‘Proposed imageprocessing pipeline and Visualiza-tion and validation’’ sections) toprototypes construction (see‘‘Animal models, Imaging modal-ities, Spatial scopes and resolu-tions, Developmental stages, Datatypes, and Matching procedures’’sections) paved the way for bio-logical insights in developmentalprocesses (Luengo-Oroz et al.,2011). Below, we comment someof the most prominent resultsderived from anatomical and geneexpression atlases.Kovacevic et al. (2005) used an

atlas model to perform geneticand anatomic phenotyping,achieving the automated detectionof mutant strains. Atlases also hadmajor implications in evolutionarystudies and Tomer et al. (2010)identified related parts of the brainin phylogenetically distant ani-mals. Chiang et al. (2011) createda comprehensive brain wiring mapof the adult Drosophila brain which



provides a crucial tool to analyzeinformation processing within andbetween neurons.Using all the genetic information

gathered in their model, Friseet al. (2010) clustered genescoexpression domains to elucidatepreviously unknown genetic func-tions and molecular and geneticinteractions. Lein et al. (2007)detected highly specific cellularmarkers and deciphered cellularheterogeneity previously unidenti-

fied in the adult mouse brain witha gene expression atlas of morethan 20,000 genes. Carson et al.(2005) discovered gene expres-sion possibly related to Parkinsondisease.Liu et al. (2009) showed in C.

elegans that different gene regula-tory pathways can correlate withidentical cell fates. However, cellfate modules with specific molecu-lar signatures repeatedly occurredalong the cell lineage, revealing

bifurcations toward cell differentia-tion. Fisher et al. (2011) identifiedthe digits identity in chick embryosby the computational analysis ofgenes expression and cell fate.The association of cell trackingtechniques (McMahon et al., 2008;Pastor et al., 2009; Suppatto etal., 2009) together with geneexpression atlases lead to fatemaps, which are promising toolsfor stem cell studies and regenera-tive medicine.

Figure 3. Image processing pipeline to build a digital atlas model: After preprocessing, nuclei detection and cell segmentation algo-rithms are applied to extract cell position and volume. This information is combined with quantification schemes. These operations,iterated throughout a cohort of individuals, yield cellular-level, quantitative measurements of many genetic and/or functional pat-terns. A common reference and/or landmarks highlighted in all individuals are automatically segmented and identified to steer aregistration procedure. The latter multiplex all measured patterns into a single, digital template. The resulting atlas can be validatedand mined through dedicated, interactive visualization tools.



PERSPECTIVES

The next generation of in situhybridization techniques is ex-pected to overcome the currentlimitations in the number of genepatterns that can be simultane-ously labeled in a single specimen.Choi et al. (2010) recently devel-oped a new multiplexing techniquethat allows to fluorescently tag upto five different mRNA targets at atime. Compared to present doubleor triple in situ hybridization tech-niques, this scheme will drasticallyfacilitate the acquisition of dataand matching operations.A long anticipated goal (Megason

and Fraser, 2007) in computationalbiology consists in the reconstruc-tion of continuous spatiotemporalprototypes where gene expressioncan be determined for every cell inthe embryo not only at certain, dis-crete developmental stages, but atany possible developmental time:(x, y, z, t, G). Achieving this goaldepends on the proper integrationof gene expression atlases with thereconstructed cell lineage tree(Supatto et al., 2009; Olivier et al.,2010; Luengo-Oroz et al., 2012b).Recent work in this direction(Castro-Gonzalez et al., 2010) indi-cates that 3D1 time atlases towardan integrated model of living sys-tems and multiscale dynamics,require the use of image process-ing techniques operating directly inthe 4D space (Luengo-Oroz et al.,2012a).Finally, a key future challenge

revolves around achieving stand-ards and making databases com-ing from different laboratoriesinteroperable. As data acquisitiongoes on at an accelerated pace, itbecomes crucial to achieve sys-tems helping to contribute, organ-ize, and find relevant data. As anexample, machine learning hasbeen recently applied to the auto-matic recognition and ontologicalannotation of gene expression pat-terns in the mouse embryo withanatomical terms (Han et al.,2011). Following this trend, therehas been a series of recent stand-ardization efforts to the crossplat-form integration of multimodaldata through the use of controlledterminologies, or ontologies (Diez-

Roux et al., 2011) that can beaccessed by query-based web sys-tems (Hawrylycz et al., 2011;Milyaev et al., 2012). This will ulti-mately allow the systematic com-parison of individuals within andeven between species.

REFERENCES

Abbott A. 2009. Microscopic marvels:seeing the system. Nature 459:630–631.

Aljabar P, Heckemann R, Hammers A,et al. 2009. Multi-atlas based seg-mentation of brain images: atlasselection and its effect on accuracy.Neuroimage 46:726–738.

Amatruda J, Shepard J, Stern H, ZonL. 2002. Zebrafish as a cancer modelsystem. Cancer Cell 1:229–231

Ashburner M, Ball C, Blake J, et al.2000. Gene ontology: tool for the uni-fication of biology. Nat Genet 25:25–29.

Baker M. 2010. Screening: the age offishes. Nat Meth 8:47–51.

Baldock R, Bard J, Burger A, et al.2003. EMAP and EMAGE. Neuroinfor-matics 1:309–325.

Barros T, Alderton W, Reynolds H,et al. 2008. Zebrafish: an emergingtechnology for in vivo pharmacologi-cal assessment to identify potentialsafety liabilities in early drug discov-ery. Br J Pharmacol 154:1400–1413.

Blanchoud S, Budirahardja Y, Naef F,Gonczy P. 2010. ASSET: a robustalgorithm for the automated seg-mentation and standardization ofearly Caenorhabditis elegansembryos. Dev Dyn 239:3285–3296.

Boline J, Lee E, Toga A. 2008. Digitalatlases as a framework for data shar-ing. Frontiers Neurosci 2:100–106.

Brend T, Holley SA. 2009. ZebrafishWhole Mount High-Resolution DoubleFluorescent In Situ Hybridization. JVis Exp 25:1229.

Bruckner S, Solteszova V, Groller M,et al. 2009. Braingazer-visual queriesfor neurobiology research. IEEETrans Visual Comput Graphics 15:1497–1504.

Carson J, Ju T, Lu H, et al. 2005. A dig-ital atlas to characterize the mousebrain transcriptome. PLoS ComputBiol 1:e41.

Castro C, Luengo-Oroz M, DesnoulezS, et al. 2009.An automatic quantifi-cation and registration strategy tocreate a gene expression atlas ofzebrafish embryogenesis. In: Pro-ceedings of IEEE EMBS Conference.pp. 1469–1472.

Castro C, Luengo-Oroz M, Douloquin L,et al. 2011.Image processing chal-lenges in the creation of spatiotempo-ral gene expression atlases of develop-ing embryos. In: Proceedings of IEEEEMBS Conference. pp. 6841–6844.

Castro-Gonzalez C, Luengo-Oroz M,Douloquin L, et al. 2010.Towards adigital model of zebrafish embryo-genesis. Integration of cell trackingand gene expression quantification.In: Proceedings of IEEE EMBS Con-ference. pp. 5520–5523.

Chiang A, Lin C, Chuang C, et al. 2011.Three-dimensional reconstruction ofbrain-wide wiring networks in Dro-sophila at single-cell resolution. CurrBiol 21:1–11.

Choi H, Chang J, Trinh L, et al. 2010.Programmable in situ amplificationfor multiplexed imaging of mRNAexpression. Nat Biotechnol 28:1208–1212.

Chudakov D, Lukyanov S, Lukyanov K.2005. Fluorescent proteins as a tool-kit for in vivo imaging. Trends Bio-technol 23:605–613.

Damle S, Hanser B, Davidson E, FraserS. 2006. Confocal quantification ofcis-regulatory reporter gene expres-sion in living sea urchin. Dev Biol299:543–550.

Davidson E, Erwin D. 2006. Gene regu-latory networks and the evolution ofanimal body plans. Science 311:796–800.

Diez-Roux G, Banfi S, Sultan M, et al.2011. A high-resolution anatomicalatlas of the transcriptome in the mouseembryo. PLoS Biol 9:e1000582.

Dorr A, Lerch J, Spring S, et al. 2008.High resolution three-dimensionalbrain atlas using an average magneticresonance image of 40 adult C57Bl/6Jmice. Neuroimage 42:60–69.

Drblikova O, Komornikova M, Remesi-kova M, et al. 2007. Estimate of thecell number growth rate using PDEmethods of image processing and timeseries analysis. J Electr Eng 58:86–92.

Evanko D, et al. 2010. Microscope har-monies. Nat Meth 7:779.

Fernandez-Gonzalez R, Munoz-BarrutiaA, Barcellos-Hoff M, Ortiz-de Solo-rzano C. 2006. Quantitative in vivo mi-croscopy: the return from the ’omics’.Curr Opin Biotechnol 17:501–510.

Fisher M, Clelland A, Bain A, et al.2008. Integrating technologies forcomparing 3D gene expressiondomains in the developing chicklimb. Dev Biol 317:13–23.

Fisher M, Downie H, Welten M, et al.2011. Comparative analysis of 3Dexpression patterns of transcriptionfactor genes and digit fate maps inthe developing chick wing. PLoS One6:e18661.

Fonseca C, Backhaus M, Bluemke D,et al. 2011. The cardiac atlas proj-ect—an imaging database for compu-tational modeling and statisticalatlases of the heart. Bioinformatics27:2288–2295.

Fowlkes C, Hendriks C, Keranen S, et al.2008. A quantitative spatiotemporalatlas of gene expression in the Dro-sophila blastoderm. Cell 133:364–374.

Frise E, Hammonds A, Celniker S.2010. Systematic image-driven anal-



ysis of the spatial Drosophila embry-onic expression landscape. Mol SystBiol 6:345.

Frolkovic P, Mikula K, Peyrieras N, SartiA. 2007. Counting number of cellsand cell segmentation using advec-tion-diffusion equations. Kybernetika43:817–829.

Giepmans B, Adams S, Ellisman M,Tsien R. 2006. The fluorescent tool-box for assessing protein locationand function. Science 312:217–224.

Goldsmith P. 2004. Zebrafish as apharmacological tool: the how, whyand when. Curr Opin Pharmacol 4:504–512.

Gorfinkiel N, Schamberg S, BlanchardG. 2011. Integrative approaches tomorphogenesis: lessons from dorsalclosure. Genesis 49:522–533.

Gouaillard A, Brown T, Bronner-FraserM, et al. 2007. GoFigure and the digi-tal fish project: open tools and opendata for an imaging based approachto system biology. Insight J. Availableat:http://hdl. handle.net/1926/565.

Han L, van Hemert J, Baldock R. 2011.Automatically identifying and anno-tating mouse embryo gene expres-sion patterns. Bioinformatics 27:1101–1107.

Hawrylycz M, Baldock R, Burger A,et al. 2011. Digital atlasing andstandardization in the mouse brain.PLoS Comput Biol 7:e1001065.

Hedges S, Dudley J, Kumar S. 2006.TimeTree: a public knowledge-baseof divergence times among organ-isms. Bioinformatics 22:2971–2972.

Hendriks C, Keranen S, Fowlkes C,et al. 2006. Three-dimensional mor-phology and gene expression in theDrosophila blastoderm at cellular re-solution I: data acquisition pipeline.Genome Biol 7:R124.

Hill A, Teraoka H, Heideman W, Peter-son R. 2005. Zebrafish as a modelvertebrate for investigating chemicaltoxicity. Toxicol Sci 86:6–19.

Huisken J, Stainier D. 2009. Selectiveplane illumination microscopy techni-ques in developmental biology. De-velopment 136:1963–1975.

Jacobs R, Papan C, Ruffins S, et al.2003. MRI: volumetric imaging forvital imaging and atlas construction.Nat Rev Mol Cell Biol 4:SS10–SS16.

Johnson G, Badea A, Brandenburg J,et al. 2010. Waxholm space: animage-based reference for coordinat-ing mouse brain research. Neuro-image 53:365–372.

Jones T, Kang I, Wheeler D, et al.2008. CellProfiler analyst: data ex-ploration and analysis software forcomplex image-based screens. BMCBioinform 9:482, doi: 10.1186/1471-2105-9-482.

Keller P, Schmidt A, Santella A, et al.2010. Fast, high-contrast imaging ofanimal development with scannedlight sheet-based structured-illumi-nation microscopy. Nat Meth 7:637–642.

Kerwin J, Yang Y, Merchan P, et al.2010. The HUDSEN atlas: a three-di-mensional (3D) spatial framework forstudying gene expression in thedeveloping human brain. J Anat 217:289–299.

Khairy K, Keller P. 2011. Reconstruct-ing embryonic development. Genesis49:488–513.

Kovacevic N, Henderson J, Chan E,et al. 2005. A three-dimensional MRIatlas of the mouse brain with esti-mates of the average and variability.Cereb Cortex 15:639–645.

Kriva Z, Mikula K, Peyrieras N, et al.2010. 3D early embryogenesis imagefiltering by nonlinear partial differen-tial equations. Med Image Anal 14:510–526.

Lau C, Ng L, Thompson C, et al. 2008.Exploration and visualization of geneexpression with neuroanatomy in theadult mouse brain. BMC Bioinform 9:153.

Lein E, Hawrylycz M, Ao N, et al. 2007.Genome-wide atlas of gene expres-sion in the adult mouse brain. Nature445:168–176.

Li E, Davidson E. 2009. Building devel-opmental gene regulatory networks.Birth Def Res Part C: Embryo Today:Rev 87:123–130.

Liu X, Long F, Peng H, et al. 2009.Analysis of cell fate from single-cellgene expression profiles in C. ele-gans. Cell 139:623–633.

Long F, Peng H, Liu X, et al. 2009. A3D digital atlas of C. elegans and itsapplication to single-cell analyses.Nat Meth 6:667–672.

Lu J, Hsia Y, Tu H, et al. 2011. Liverdevelopment and cancer formation inzebrafish. Birth Def Res Part C:Embryo Today: Rev 93:157–172.

Luengo-Oroz M, Duloquin L, Castro C,et al. 2008.Can voronoi diagrammodel cell geometries in earlysea-urchin embryogenesis? In: Pro-ceedings of IEEE ISBI Conference. pp.504–507.

Luengo-Oroz M, Ledesma-Carbayo M,Peyrieras N, Santos A. 2011. Imageanalysis for understanding embryodevelopment: a bridge from micros-copy to biological insights. Curr OpinGenet Dev 21:630–637.

Luengo-Oroz M, Pastor-Escuredo D,Castro-Gonzalez C, et al. 2012a.3D1t morphological processing:applications to embryogenesis imageanalysis. IEEE Trans Image Process.In press (doi: 10.1109/TIP.2012.2197007).

Luengo-Oroz M, Rubio-Guivernau J,Faure E, et al. 2012b. Methodologyfor reconstructing early zebrafish de-velopment from in-vivo multiphotonmicroscopy. IEEE Trans Image Pro-cess 21:2335–2340.

Ma Y, Hof P, Grant S, et al. 2005. Athree-dimensional digital atlas data-base of the adult C57BL/6J mousebrain by magnetic resonance micros-copy. Neuroscience 135:1203–1215.

MacKenzie-Graham A, Lee E, Dinov I,et al. 2004. A multimodal, multidi-mensional atlas of the C57BL/6Jmouse brain. J Anat 204:93–102.

Maintz J, Viergever M. 1998. A surveyof medical image registration. MedImage Anal 2:1–36.

Martone M, Zaslavsky I, Gupta A, et al.2008. The smart atlas: spatial andsemantic strategies for multiscaleintegration of brain data. In: Anat-omy Ontologies for Bioinformatics,Computational Biology, Volume 6,Part III, 267–286, DOI: 10.1007/978-1-84628-885-2_13.

Mazziotta J, Toga A, Evans A, et al. 2001.A probabilistic atlas and reference sys-tem for the human brain: internationalconsortium for brain mapping (ICBM).Philos Trans R Soc London Ser B: BiolSci 356:1293–1322.

McMahon A, Supatto W, Fraser S, Sta-thopoulos A. 2008. Dynamic analysesof drosophila gastrulation provideinsights into collective cell migration.Science 322:1546–1550.

McPherson J, Marra M, Hillier L, et al.2001. A physical map of the humangenome. Nature 409:934–941.

Megason S, Fraser S. Imaging in systemsbiology. 2007. Cell 130:784–795.

Mikula K, Peyrieras N, Remesıkova M,Stasova O. 2011. Segmentation of3D cell membrane images by PDEmethods and its applications. Com-put Biol Med 41:326–339.

Milan D, Peterson T, Ruskin J, et al.2003. Drugs that induce repolariza-tion abnormalities cause bradycardiain zebrafish. Circulation 107:1355–1358.

Milyaev N, Osumi-Sutherland D, ReeveS, et al. 2012. The virtual fly brainbrowser and query interface. Bioin-formatics, 28:411–415.

Murray J, Bao Z, Boyle T, et al. 2008.Automated analysis of embryonic geneexpression with cellular resolution inC. elegans. Nat Meth 5:703–709.

Ng L, Bernard A, Lau C, et al. 2009. Ananatomic gene expression atlas ofthe adult mouse brain. Nat Neurosci12:356–362.

Ng L, Pathak S, Kuan C, et al. 2007.Neuroinformatics for genome-wide3D gene expression mapping in themouse brain. IEEE/ACM Trans Com-put Biol Bioinform 4:382–393.

North T, Goessling W, Walkley C, et al.2007. Prostaglandin e2 regulates ver-tebrate haematopoietic stem cell ho-meostasis. Nature 447:1007–1011.

Oates A, Gorfinkiel N, Gonzalez-GaitanM, Heisenberg C. 2009. Quantitativeapproaches in developmental biol-ogy. Nat Rev Genet 10:517–530.

Olivier N, Luengo-Oroz M, Duloquin L,et al. 2010. Cell lineage reconstruc-tion of early zebrafish embryos usinglabel-free nonlinear microscopy. Sci-ence 329:967–971.

Pardo-Martin C, Chang T, Koo B, et al.2010. High-throughput in vivo verte-brate screening. Nat Meth 7:634–636.



Park H, Bland P, Meyer C. 2003. Con-struction of an abdominal probabilis-tic atlas and its application in seg-mentation. IEEE Trans Med Imag 22:483–492.

Pastor D, Luengo-Oroz M, Lombardot B,et al. 2009.Cell tracking in fluorescenceimages of embryogenesis processeswith morphological reconstruction by4D-tubular structuring elements. In:Proceedings of IEEE EMBS Conference.pp. 970–973.

Paul B, Hong C, Sachidanandan C,et al. 2008. Dorsomorphin inhibitsbmp signals required for embryogen-esis and iron metabolism. Nat ChemBiol 4:33–41.

Peng H. 2008. Bioimage informatics: anew area of engineering biology. Bio-informatics 24:1827.

Peng H, Chung P, Long F, et al. 2011.Brainaligner: 3D registration atlasesof drosophila brains. Nat Meth 8:493–498.

Peng H, Long F, Liu X, et al. 2008.Straightening Caenorhabditis elegansimages. Bioinformatics 24:234–242.

Peng H, Ruan Z, Long F, et al. 2010.V3D enables real-time 3D visualiza-tion and quantitative analysis oflarge-scale biological image datasets. Nat Biotechnol 28:348–353.

Pisarev A, Poustelnikova E, SamsonovaM, Reinitz J. 2009. Flyex, the quanti-tative atlas on segmentationgene expression at cellular resolu-tion. Nucl Acids Res 37 (Suppl 1):D560–D566.

Potikanond D, Verbeek F. 2011. 3Dvisual integration of spatio-temporalgene expression patterns on digitalatlas of zebrafish embryo using webservice. In: Proceedings of Interna-tional Conference on Advances inCommunication and InformationTechnology. pp. 56–62.

Qu L, Peng H. 2010. A principal skele-ton algorithm for standardizing con-focal images of fruit fly nervous sys-tems. Bioinformatics 26:1091–1097.

Rex D, Ma J, Toga A. 2003. The lonipipeline processing environment.Neuroimage 19:1033–1048.

Richardson L, Venkataraman S, Ste-venson P, et al. 2010. Emage mouseembryo spatial gene expressiondatabase: 2010 update. Nucl AcidsRes 38 (Suppl 1):D703–D709.

Rittscher J, Padfield D, Santamaria A,et al. 2011.Methods and algorithmsfor extracting high-content signa-tures from cells, tissues, and model

organisms. In: Proceedings of IEEEISBI Conference. pp. 1712–1716.

Rittscher J, Yekta A, Musodiq O, et al.2010. Automated systems and meth-ods for screening zebrafish. US Pat-ent NO.2010/0119119 A1.

Rubel O, Weber G, Huang M, et al.2010. Integrating data clustering andvisualization for the analysis of 3Dgene expression data. IEEE/ACMTrans Comput Biol Bioinform 7:64–79.

Rubio-Guivernau J, Gurchenkov V,Luengo-Oroz M, et al. 2012. Wave-let-based image fusion in multiviewthree-dimensional microscopy. Bioin-formatics 28:238–245.

Ruffins SW, Martin M, Keough L, et al.2007. Digital three-dimensional atlasof quail development using high-re-solution MRI. Sci World J 7:592–604.

Sharpe J, Ahlgren U, Perry P, et al.2002. Optical projection tomographyas a tool for 3D microscopy and geneexpression studies. Science 296:541–545.

Sipes N, Padilla S, Knudsen T. 2011.Zebrafish as an integrative model fortwenty-first century toxicity testing.Birth Def Res C: Embryo Today: Rev93:256–267.

Smith S, Jenkinson M, Woolrich M,et al. 2004. Advances in functionaland structural MR image analysis andimplementation as FSL. Neuroimage23:S208–S219.

Supatto W, McMahon A, Fraser S, Sta-thopoulos A. 2009. Quantitativeimaging of collective cell migrationduring drosophila gastrulation: multi-photon microscopy and computationalanalysis. Nat Protocol 4:1397–1412.

Toga A, Thompson P, Mori S, et al.2006. Towards multimodal atlases ofthe human brain. Nat Rev Neurosci7:952–966.

Tomer R, Denes A, Tessmar-Raible K,Arendt D. 2010. Profiling by imageregistration reveals common origin ofannelid mushroom bodies and verte-brate pallium. Cell 142:800–809.

Truong T, Supatto W. 2011. Towardhigh-content/high-throughput imag-ing and analysis of embryonic mor-phogenesis. Genesis 49:555–569.

Ullmann J, Cowin G, Kurniawan N, Col-lin S. 2010. A three-dimensional digi-tal atlas of the zebrafish brain. Neu-roimage 51:76–82.

Vonesch C, Aguet F, Vonesch J, UnserM. 2006. The colored revolution ofbioimaging. IEEE Signal Process Mag23:20–31.

Walter G, Bussow K, Lueking A, Glokler J.2002. High-throughput protein arrays:prospects for molecular diagnostics.Trends Mol Med 8:250–253.

Walter T, Shattuck D, Baldock R, et al.2010. Visualization of image datafrom cells to organisms. Nat Meth 7:S26–S41.

Warga R, Kane D. 2003. One-eyed pin-head regulates cell motility inde-pendent of squint/cyclops signaling.Dev Biol 261:391–411.

Weber G, Rubel O, Huang M, et al.2009. Visual exploration of three-di-mensional gene expression usingphysical views and linked abstractviews. In: IEEE/ACM Transactions onComputational Biology and Bioinfor-matics. 5:296–309.

Welten M, de Haan S, van den BoogertN, et al. 2006. ZebraFISH: fluores-cent in situ hybridization protocoland three-dimensional imaging ofgene expression patterns. Zebrafish3:465–476.

Woods R, Dapretto M, Sicotte N, et al.1999. Creation and use of a Talair-ach-compatible atlas for accurate,automated, nonlinear intersubjectregistration, and analysis of func-tional imaging data. Hum Brain Map8:73–79.

Woods R, Grafton S, Watson J, et al.1998. Automated image registration.II. Intersubject validation of linearand nonlinear models. J ComputAssist Tomogr 22:153–165.

Wu J, Pollard T. 2005. Counting cytoki-nesis proteins globally and locally infission yeast. Science 310:310–314.

Yang L, Ho N, Alshut R, et al. 2009.Zebrafish embryos as models forembryotoxic and teratological effectsof chemicals. Reprod Toxicol 28:245–253.

Zanella C, Campana M, Rizzi B, et al.2010. Cells segmentation from 3-Dconfocal images of early zebrafishembryogenesis. IEEE Trans ImageProcess 19:770–781.

Zaslavsky I, He H, Tran J, et al.2004.Integrating brain data spa-tially: spatial data infrastructureand atlas environment for onlinefederation and analysis of brainimages. In: IEEE Proceedings ofInternational Workshop on Databaseand Expert Systems Applications. pp.389–393.

Zitova B, Flusser J. 2003. Image regis-tration methods: a survey. Image VisComput 21:977–1000.



assembling models of embryo development: image …assembling models of embryo development: image...

Documents