REVIEW

Toward an Unbiased Evolutionary Platform for UnravelingXenopus Developmental Gene Networks

Ronny Beer,1 Florian Wagner,1 Vladislav Grishkevich,1 Leonid Peshkin,2 and Itai Yanai1*

1Department of Biology, Technion—Israel Institute of Technology, Haifa 32000, Israel

2Department of Systems Biology, Harvard Medical School, Boston, Massachusetts

Received 28 July 2011; Accepted 21 September 2011

Summary: The availability of both the Xenopus tropica-lis genome and the soon to be released Xenopus laevisgenome provides a solid foundation for Xenopus devel-opmental biologists. The Xenopus community has pres-ently amassed expression data for �2,300 genes in theform of published images collected in the Xenbase, theprincipal Xenopus research database. A few of thesegenes have been examined in both X. tropicalis and X.laevis and the cross-species comparison has been pro-ven invaluable for studying gene function. A recentlypublished work has yielded developmental expressionprofiles for the majority of Xenopus genes across four-teen developmental stages spanning the blastula, gas-trula, neurula, and the tail-bud. While this data was orig-inally queried for global evolutionary and developmen-tal principles, here we demonstrate its general use forgene-level analyses. In particular, we present theaccessibility of this dataset through Xenbase anddescribe biases in the characterized genes in terms ofsequence and expression conservation across the twospecies. We further indicate the advantage of examin-ing coexpression for gene function discovery relating todevelopmental processes conserved across species.We suggest that the integration of additional large-scale datasets—comprising diverse functional data—into Xenbase promises to provide a strong foundationfor researchers in elucidating biological processesincluding the gene regulatory programs encodingdevelopment. genesis 50:186–191, 2012. VVC 2011 Wiley

Periodicals, Inc.

INTRODUCTION

Xenopus is an efficient vertebrate model for geneticand developmental research benefitting from severaladvantages, including relatively affordable transgenesis,

when compared to zebrafish and mice (Hirsch et al.,2002). The early embryos are amenable to manipula-tion, thus enabling efficient microsurgery and microin-jection (Sive et al., 2000). Historically, Xenopus laevis

was the exclusive Xenopus species to be used, but overthe past two decades, Xenopus tropicalis has emergedas an excellent experimental model with powerful com-munity resources (Kashiwagi et al., 2010).

While both X. tropicalis and X. laevis offer rapid ratesof embryogenesis and large spawning, each carries dis-tinct advantages and disadvantages (Chalmers et al.,2005). The two species are easy to distinguish, since X.

laevis is characterized by significantly larger embryos andadults, as well as more robust eggs and better husbandrypractices. However, X. laevis poses a substantial setbackas a genetic system due to an allotetraploid genome withtwo related paralogues for most genes (Hellsten et al.,2007). X. tropicalis has a shorter generation time of 3–4.5 months, a potential for much greater spawning num-ber (1,000–9,000 vs. 300–1,000), and is the only knownXenopus species with a diploid genome, which is almosthalf the size of that of X. laevis (Hirsch et al., 2002). How-ever, both X. tropicalis and X. laevis exhibit a nearlyindistinguishable embryonic development (apart fromthe size differences) and adult and larval morphologiesdespite having undergone 30–90 million years of inde-pendent evolution (Evans et al., 2004).

Additional Supporting Information may be found in the online version

of this article.

* Correspondence to: Itai Yanai, Department of Biology, Technion –

Israel Institute of Technology, Haifa 32000, Israel.

E-mail: yanai@technion.ac.il

Published online 28 September 2011 in

Wiley Online Library (wileyonlinelibrary.com).

DOI: 10.1002/dvg.20811

' 2011 Wiley Periodicals, Inc. genesis 50:186–191 (2012)

Another advantage of X. tropicalis is the availabilityof a fully sequenced genome. While the full X. laevis ge-nome is still awaiting publication, thousands of X. lae-vis mRNA and EST sequences are readily available. Inour comparison of the coding regions of X. laevis andX. tropicalis, we found the sequences to be on average90% identical (Fig. 1). In comparison, the X. laevis

paralogs resulting from the genome duplication aremore closely related to each other than the orthologs(Fig. 1). This provides an intermediate genetic distancebetween such models as human–mouse (85%) andhuman–chimp (99%) (Consortium, 2005; Waterstonet al., 2002). In this brief review, we consider the com-parative study of this species-pair as a research tool tounderstand developmental and evolutionary processes.

Whole Transcriptomic Comparisons BetweenX. tropicalis and X. laevis

Comparing the two Xenopus species has long beenthought to be useful for unraveling conserved develop-mental processes. The first study to compare geneexpression in whole transcriptomic fashion invoked aDNA microarray designed for X. tropicalis genes andhybridized with RNA from both X. tropicalis and X. lae-

vis (Chalmers et al., 2005). RNA from Stages 3 (four-cellstage) and 19 (neurula) were compared and the overlapin gene expression identified using X. laevis and X. tro-

picalis RNA was shown to be congruent. This dataallowed for more efficient access to comparative geneexpression data without the trouble of obtaining X. lae-

vis-specific resources.A similarly designed study focused on transcriptomes

of Stages 10 and 40 as well as the egg, ovary, and liverand found overall congruent changes in expressionacross species (Sartor et al., 2006). Notably, this studyalso reported a correlation between the sequence diver-

gence and expression divergence of the two species(also see below). However, this method was limited byan inability to distinguish between technical variationdue to faulty hybridization and gene expression diver-gence across species. Consequently, single-speciesmicroarrays have been called into question in compara-tive transcriptomics (Chain et al., 2008). In contrast tothe previous studies, this study compared two allotetra-ploid species (X. laevis and X. borealis) which mayhave introduced an additional confounding effect.

Recently, a study including some of us set out to com-pare the developmental transcriptomes of X. laevis vs.X. tropicalis (Yanai et al., 2011). In this study, wedesigned two custom-microarrays manufactured by Agi-lent that are species-specific to X. laevis and X. tropica-

lis, respectively. To enable an appropriate comparison,the corresponding 60-mer probes were designed as vir-tually identical (>90%), solely adapted to the respectivegenomes, and three probes were assigned to eachortholog pair. We were interested in comparing the tim-ing and level of gene expression between the two spe-cies across development. We found that the timings ofgene expression upregulations across species are verysimilar for most of the genes examined (i.e., little het-erochrony). However, significant variation in expressionlevels (heterometry) was found to a much greaterextent than heterochrony. Interestingly, differences inthe levels of gene expression were concentrated in theearliest embryonic stages and declined during embryo-genesis. Furthermore, specific pathways related to theorganisms’ response to the environment were found tobe associated with coherent changes in the timing ofthe onset of expression. In addition to these global anal-yses, the expression plots for individual genes are ofinterest to those comparing the expression across thedevelopment of the two species.

Comparison of the Expression Profiles ofDevelopmental Genes

Gene expression profiles have been previously com-pared for several developmentally important genes, andhere we review these for the neurogenin and nogginfamilies. The neurogenins comprise the highly con-served family of basic helix-loop-helix (bHLH) pro-teins—neurog1, neurog2, and neurog3—essential forneuronal determination and cell-type specification (Maet al., 1996). The three originated from subsequentgene duplications along vertebrate evolution. The corre-sponding neurogenin genes from the two species share75–89% overall amino acid identity as compared to a45–57% identity with mouse sequences (Nieber et al.,2009).

Figure 2a shows in situ expression profiles from arecent work (Nieber et al., 2009) across both X. laevis

and X. tropicalis. While analyzing the expression profile

FIG. 1. Identity between the coding sequences of Xenopus laevisand Xenopus tropicalis (light grey) and between Xenopus laevis paral-ogs (dark grey). Sequence identity was determined using blastnbetween the 11,095 pairs of the previously delineated orthologs (Yanaiet al., 2011). Pairs of X. laevis paralogs were retrieved from a recentstudy (Hellsten et al., 2007) and compared as for the orthologs.

187COMPARATIVE TRANSCRIPTOMICS FOR XENOPUS DEVELOPMENT

of neurog1 in both X. laevis and X. tropicalis, it wasreported that expression was present throughout em-bryonic development upon initial induction and gastru-lation. Notably, expression appeared stronger in X. tro-

picalis, which was attributed to ‘‘relative flexibility inexpression levels’’ as well as the variation in sizes ofthese two embryos. Figure 2b indicates the temporalexpression profile identified by the developmental time-course comparison. The two different kinds of imagescharting the expression level differences highlight thedifferences seen in the in situ: (1) X. tropicalis showshigher expression levels, and (2) there appears to be aslight heterochronic shift to earlier expression in X. lae-

vis, which is hinted in the early stage in situ as well.Another comparative study of the gene expression

profiles across X. laevis and X. tropicalis addressednoggin genes (Fletcher et al., 2004), a family of secretedproteins important for patterning throughout the devel-opment of the vertebrates. It was first identified in thecDNA derived from an X. laevis embryo (Smith and Har-land, 1992). As for the neurogenin family, in X. laevis,noggin is induced and consequently expressed through-out development. The X. tropicalis and X. laevis nog-

gin orthologs share a 98% amino acid identity as well asa very similar pattern of expression (Fig. 2c). Noggin2is a bone morphogenic protein antagonist shown to be

crucial for the induction of dorsal structures in theembryo. A comparison revealed similar expression pro-files (Eroshkin et al., 2006) also evidenced by the devel-opmental time-course data (Fig. 2d).

Introduction of transcriptome profiles to Xenbase

Xenbase is an online database to aid in the study of X.laevis and X. tropicalis biology and genomics (Boweset al., 2010). It is the main resource for developmentand genomic information in the Xenopus community,continuously integrating research from the literature,including images relevant to gene expression, anatomyand developmental stages. As of 2011, Xenbase con-tained 2,316 genes with at least one expression imageof the kind shown in Figure 2a for the neurog1 gene.We submitted the 11,095 expression plots of orthologs(Yanai et al., 2011) to Xenbase and we now report theirintegration. For example, the expression images forneurog1 now also include the time-course profile (Fig.2b). Thus, such a gene that was already associated withpublished expression images is now complementedwith the temporal expression plots. Moreover, �80% ofthe genes for which we submitted the plots are nowassociated with their first expression image.

FIG. 2. Comparative gene expression profiles across Xenopus laevis and Xenopus tropicalis. a. In situ gene expression comparison forneurog1 as determined in Nieber et al. (2009). Numbers indicate developmental stages. Top row is X. tropicalis. This image is taken from(Nieber et al., 2009). b. Gene expression comparison from the developmental time-course microarray data (Yanai et al., 2011). Fourteenstages were sampled and the units of expression are in log10 scale. c,d. Expression profiles for noggin and noggin2 in the same format as b.

188 BEER ET AL.

To appraise how beneficial the additional expressionimages are for the community we asked if detectablebiases existed in terms of which genes were previouslyassociated with expression images in Xenbase. We firstlooked at sequence similarity between X. laevis and X.

tropicalis orthologs (Fig. 1). For each set of genes witha given sequence similarity (for example 90% DNAsequence identity), we computed the fraction that hadcontained expression images. We found that thoseorthologs with a higher sequence similarity featured ahigher fraction of genes with expression images (Fig.3a). Because sequence conservation is a proxy for func-tional importance (via selective pressures), the trendshown in Figure 3a can be seen as a demonstration ofthe Xenopus community’s prioritization on functionallyimportant genes. This is likely a result of the method bywhich genes were isolated in the pregenomic era,where conservation was required by low-stringencyhybridizations with genes from distant species.

Next, we examined the distribution of genes withexpression images for genes with different levels ofexpression conservation across the two species. Anindex for computing expression distance was intro-duced (Yanai et al., 2011) and computes the Euclideandistance between plots. For example, neurog1 exhibitsmore expression divergence across the species thannoggin or noggin2 (Fig. 2b–d). We partitioned thegenes in equally populated bins of increasing expres-sion similarity. For each bin we computed the fractionof genes with expression images. We found that thosegenes with high expression similarity show a higherfraction of genes with expression images (Fig. 3b). Aswith sequence similarity, this correlation likely reflects

the notion that genes under stronger selection are bothevolving slower at the expression level as well as play-ing a more important functional role and consequentlyunder closer study according to the information foundin Xenbase. Overall, the biases shown in Figure 3 indi-cate that researchers have predictably focused on themore important genes as judged by their expressionand sequence conservation. However, even for thisclass of genes, thousands have lacked images (Fig. 3) forwhich there are now temporal expression data.

Finally, we reexamined the relationship betweenexpression divergence and sequence divergence (Sartoret al., 2006) by comparing the expression divergence ofthe orthologous profiles with the sequence similarity.We found a very weak yet significant correlationbetween the two (R 5 0.08, P < 10213). Thus, overall,the selective forces acting on the evolution of sequenceand expression are intertwined, though to a limiteddegree. Perhaps more extensive work on characterizingboth the spatial and temporal expression will moreaccurately gauge expression divergence.

One important aspect of the transcriptomics datarelates to the issue of expression of gene duplicates, aproblem that is made worse in X. laevis which sus-tained a genome duplication following its last commonancestor with X. tropicalis. Because of the lack of acompleted genome for X. laevis and the difficult inresolving the expression of genes with near identicalsequences using expression microarrays, there was aninherent problem in measuring expression of recentduplicates. In light of this, we adopted a strategy ofselecting probes in conserved sequence regions suchthat the resulting expression of duplicates may be repre-

FIG. 3. Biases of Xenopus research towards genes with higher conservation of sequence and expression. a. Genes with Xenbase expres-sion images as a function of their sequence similarity. The plot indicates the fraction of genes with expression images in Xenbase for theshown sequence similarity ranges. b. Genes with Xenbase expression images as a function of their expression similarity. Expression diver-gences between X. laevis and X. tropicalis (Yanai et al., 2011) were sorted and split into equally populated bins. The fraction of genes ineach bin with Xenbase images is shown.

189COMPARATIVE TRANSCRIPTOMICS FOR XENOPUS DEVELOPMENT

sented as the sum of their separate expressions. Thus,many interesting patterns of subfunctionalization arepotentially masked in our profiles for duplicates, asrecently revealed (Hellsten et al., 2007).

Gene Expression Clustering Toward IdentifyingDevelopmental Networks

Beyond examining the expression profile of particulargenes, the temporal expression profiles can be used toidentify new genes participating in specific pathways,based upon an operational assumption that genes withsimilar expression profiles are functionally related. Givena gene of interest, for examplemix1, the expression pro-file can be analyzed and located in a specific cluster (Fig.4a). Upon analyzing this cluster, other genes with similarexpression profiles are identified, such as nodal1. Inter-estingly, the nodal1 gene may be functionally related tomix1 through its close paralog mix2 (Cao et al., 2008).Additionally, the well-studied gene brachyury1.1 is also amember of this cluster. All three have been studied andthus, known information can be applied to other, per-haps lesser known or unknown, genes in this cluster; forexample, LOC100489561, currently only annotated as

‘‘sbk1-like.’’ Furthermore, sbk1 is also a member of thiscluster (Fig. 4a), and these two genes with very similarexpression profiles may indicate gene duplicates withconserved expression.

Notably, goosecoid is also known to be a part of thisnetwork of genes; however, it was not clustered withthem. Rather goosecoid is in another cluster with a simi-lar expression profile, particularly marked by upregula-tion at Stage 9 (Fig. 4b). Following this upregulation, thelevel of expression decreases at a slower rate than thepreviously noted cluster (Fig. 4a). This suggests thatwhen studying individual gene networks, a more appro-priate clustering strategy might involve restricting theanalysis to those stages in which this network is induced.This analysis highlights the important role of computa-tional analysis in developmental transcriptome data. Fig-ure 4c,d shows two additional expression clusters inwhich expression is often not particularly well conservedacross the two species. This lack of conservation also pro-vides important clues as to the functional component of agene’s expression profiles (Yanai and Hunter, 2009).

To help in this search we developed a platform-inde-pendent gene expression browser using the MATLABenvironment. This browser provides a simple graphical

FIG. 4. Clustering of comparative gene expression profiles. a–d. Each plot indicates a plot of genes clustered by Heyer clustering as pre-sented previously (Yanai et al., 2011). The clustering was performed only on X. laevis profiles. Four expression clusters are shown. Theunbroken lines correspond to the X. laevis profiles while the dashed lines show the superimposed X. tropicalis profiles. The full clustering isprovided in Supporting Information Table S1.

190 BEER ET AL.

user interface-based access to the expression data. Inaddition to forward and backward advance options,genes can be accessed in random order by their EnsemblID. Expression level is displayed as raw, log-ed, or nor-malized. Each species is displayed at its own range ofexpression values to compare clutches to each other.Additionally both species and all clutches/probes arebrought together to a common scale for comparison. Inparticular, this browser has the useful ability to load anysub-set of genes and efficiently click through them. Thisbrowser is available for download here http://kirschner.med.harvard.edu/Xenopus_Transcriptomics.html.

In conclusion, we have integrated a large-scale data-set for general consumption by the Xenopus commu-nity through the widely used Xenbase portal. Thisallows for easy accessibility of the dataset and facilitatesthe research of new—perhaps less sequence andexpression conserved—genes by the community. Wesuggest that integration of additional datasets such asprotein expression levels, protein–protein interactions,and gene regulatory linkages, will further help research-ers in unraveling the underlying developmental genenetworks in the Xenopus embryo.

ACKNOWLEDGMENTS

The authors thank the Xenbase team, in particularKevin Snyder, for their inclusion of the developmentaltime-course plots. They also thank Tamar Hashimshonyfor a critical reading of the manuscript. They also thankthe two anonymous reviewers for valuable suggestions.

LITERATURE CITED

Bowes JB, Snyder KA, Segerdell E, Jarabek CJ, Azam K,Zorn AM, Vize PD. 2010. Xenbase: Gene expressionand improved integration. Nucleic Acids Res38:D607–D612.

Cao Y, Siegel D, Oswald F, Knochel W. 2008. Oct25represses transcription of nodal/activin target genesby interaction with signal transducers during Xeno-pus gastrulation. J Biol Chem 283:34168–34177.

Chain FJ, Ilieva D, Evans BJ. 2008. Single-species micro-arrays and comparative transcriptomics. PLoS One3:e3279.

Chalmers AD, Goldstone K, Smith JC, Gilchrist M,Amaya E, Papalopulu N. 2005. A Xenopus tropicalisoligonucleotide microarray works across speciesusing RNA from Xenopus laevis. Mech Dev122:355–363.

Consortium CSaA. 2005. Initial sequence of the chim-panzee genome and comparison with the human ge-nome. Nature 437:69–87.

Eroshkin FM, Ermakova GV, Bayramov AV, Zaraisky AG.2006. Multiple noggins in vertebrate genome: Clon-ing and expression of noggin2 and noggin4 in Xeno-

pus laevis. Gene Expr Patterns 6:180–186.

Evans BJ, Kelley DB, Tinsley RC, Melnick DJ, CannatellaDC. 2004. A mitochondrial DNA phylogeny of Afri-can clawed frogs: Phylogeography and implicationsfor polyploid evolution. Mol Phylogenet Evol33:197–213.

Fletcher RB, Watson AL, Harland RM. 2004. Expressionof Xenopus tropicalis noggin1 and noggin2 in earlydevelopment: Two noggin genes in a tetrapod.Gene Expr Patterns 5:225–230.

Hellsten U, Khokha MK, Grammer TC, Harland RM,Richardson P, Rokhsar DS. 2007. Accelerated geneevolution and subfunctionalization in the pseudote-traploid frog Xenopus laevis. BMC Biol 5:31.

Hirsch N, Zimmerman LB, Grainger RM. 2002. Xenopus,the next generation: X. tropicalis genetics andgenomics. Dev Dyn 225:422–433.

Kashiwagi K, Kashiwagi A, Kurabayashi A, Hanada H,Nakajima K, Okada M, Takase M, Yaoita Y. 2010.Xenopus tropicalis: An ideal experimental animal inamphibia. Exp Anim 59:395–405.

Ma Q, Kintner C, Anderson DJ. 1996. Identification ofneurogenin, a vertebrate neuronal determinationgene. Cell 87:43–52.

Nieber F, Pieler T, Henningfeld KA. 2009. Comparativeexpression analysis of the neurogenins in Xenopus

tropicalis and Xenopus laevis. Dev Dyn 238:451–458.

Sartor MA, Zorn AM, Schwanekamp JA, Halbleib D, Kar-yala S, Howell ML, Dean GE, Medvedovic M, Tomlin-son CR. 2006. A new method to remove hybridiza-tion bias for interspecies comparison of global geneexpression profiles uncovers an associationbetween mRNA sequence divergence and differen-tial gene expression in Xenopus. Nucleic Acids Res34:185–200.

Sive HL, Grainger RM, Harland RM. 2000. Early develop-ment of Xenopus laevis: A laboratory manual, Vol.9. Cold Spring Harbor, NY: Cold Spring Harbor Labo-ratory Press. pp338.

Smith WC, Harland RM. 1992. Expression cloning ofnoggin, a new dorsalizing factor localized to theSpemann organizer in Xenopus embryos. Cell70:829–840.

Waterston RH, Lindblad-Toh K, Birney E, Rogers J, AbrilJF, Agarwal P, Agarwala R, Ainscough R, Alexanders-son M, An P, et al. 2002. Initial sequencing andcomparative analysis of the mouse genome. Nature420:520–562.

Yanai I, Hunter CP. 2009. Comparison of diverse devel-opmental transcriptomes reveals that coexpressionof gene neighbors is not evolutionarily conserved.Genome Res 19:2214–2220.

Yanai I, Peshkin L, Jorgensen P, Kirschner MW. 2011.Mapping gene expression in two Xenopus species:Evolutionary constraints and developmental flexibil-ity. Dev Cell 20:483–496.

191COMPARATIVE TRANSCRIPTOMICS FOR XENOPUS DEVELOPMENT