Charting Brachyury-mediated developmental pathwaysduring early mouse embryogenesisMacarena Lolasa,b, Pablo D. T. Valenzuelab, Robert Tjiana,c,1, and Zhe Liud,1

dJunior Fellow Program, aJanelia Farm Research Campus, Howard Hughes Medical Institute, Ashburn, VA 20147; bFundación Ciencia para la Vida, Santiago7780272, Chile; and cLi Ka Shing Center for Biomedical and Health Sciences, California Institute for Regenerative Medicine Center of Excellence, Departmentof Molecular and Cell Biology, University of California, Berkeley, CA 94720

Contributed by Robert Tjian, February 11, 2014 (sent for review January 14, 2014)

To gain insights into coordinated lineage-specification and mor-phogenetic processes during early embryogenesis, here we reporta systematic identification of transcriptional programs mediatedby a key developmental regulator—Brachyury. High-resolutionchromosomal localization mapping of Brachyury by ChIP sequenc-ing and ChIP-exonuclease revealed distinct sequence signaturesenriched in Brachyury-bound enhancers. A combination of genome-wide in vitro and in vivo perturbation analysis and cross-speciesevolutionary comparison unveiled a detailed Brachyury-depen-dent gene-regulatory network that directly links the function ofBrachyury to diverse developmental pathways and cellular house-keeping programs. We also show that Brachyury functions pri-marily as a transcriptional activator genome-wide and that anunexpected gene-regulatory feedback loop consisting of Brachyury,Foxa2, and Sox17 directs proper stem-cell lineage commitmentduring streak formation. Target gene and mRNA-sequencing cor-relation analysis of the Tc mouse model supports a crucial role ofBrachyury in up-regulating multiple key hematopoietic and muscle-fate regulators. Our results thus chart a comprehensive map ofthe Brachyury-mediated gene-regulatory network and how itinfluences in vivo developmental homeostasis and coordination.

primitive streak | early development | mesoendoderm differentiation

In the past decade, significant insights have been gained inunderstanding gene-regulatory programs responsible for em-

bryonic stem (ES) cell pluripotency and self-renewal. However,transcriptional control mechanisms underlying finely balancedlineage-segregation and morphogenetic processes during earlymammalian embryogenesis remained elusive. During early mousegastrulation, Brachyury (T), a classical enhancer-binding transcrip-tion factor (TF), has been reported to be required for the properdevelopment of primitive streak, allantois, axial, and posteriormesoderm (reviewed in ref. 1).Specifically, mouse embryos that are homozygous for the

Brachyury (T) deletion die at midgestation (2). T−/− mutantepiblast cells are compromised in their ability to migrate awayfrom the primitive streak and, therefore, are unable to undergothe morphogenetic movements carried out by their wild-type(WT) counterparts during gastrulation. In addition to defects inthe primitive streak, the notochord is absent in posterior portionsof the embryo. Although the anterior portions of T mutant micecontain notochordal precursor-like cells, they fail to undergonormal terminal differentiation (3, 4). The embryonic patternposterior to the forelimb region of T−/− animals is also disturbed,with somites posterior to the seventh pair absent or abnormal.Although neural folds fuse to form the neural tube, they areseverely kinked in the caudal region, and the surface ectodermtends to form fluid-filled blisters (2, 4, 5). In addition to thesewell-documented defects, numerous other phenotypic abnormali-ties have been reported in T−/− and T mutant embryos, includingleft–right patterning defects and morphological abnormalities inheart development (6, 7). Together, these genetic and phenotypicanalyses strongly suggest that Brachyury and Brachyury-expressing

cells play diverse and indispensable roles in early mousedevelopment.In mouse ES cell-based differentiation systems, Brachyury is

widely used as a mesoendoderm marker, and Brachyury-positivecells were able to differentiate into mesodermal and definitiveendodermal lineages, such as cardiomyocytes and hepatocytes(8–10). In a previous study, we found that depletion of a corepromoter factor, the TATA binding protein-associated factor 3,in ES cells leads to significant up-regulation of Brachyury andmesoderm lineages during ES cell differentiation (11). Here, wesystematically characterized the molecular function of Brachyuryduring ES cell differentiation by genomic, single-cell imaging,and biochemical approaches. We then contrasted our resultswith published zebrafish and Xenopus Brachyury binding sitemapping data (12, 13) to compare evolutionarily convergent ordivergent pathways. We further extended these studies to examinethe direct impact of Brachyury loss of function during early de-velopment with the Tc mouse model. Together, our data providea systematic and comprehensive view of Brachyury-mediatedregulation and also reveal unique mechanistic insights into howthis gene network likely contributes to early mouse embryogenesis.

Results and DiscussionCharacterization of in Vitro Primitive Streak Induction. We used thewell-established Activin-A–mediated differentiation protocol thatefficiently drives ES cells to form primitive streak cells in vitro (9,14) (SI Appendix, Fig. S1A). We successfully generated twohighly specific polyclonal antibodies directed against the Brachyuryprotein. At day 4 after differentiation, we detected stronginduction of primitive streak markers (Brachyury and Gsc)(SI Appendix, Figs. S1 B–D and S2A), indicating a successfuldifferentiation of primitive streak cells. To comprehensively

Significance

The gene-regulatory mechanisms for finely balanced cell-fatedetermination and morphogenesis during early animal de-velopment remain largely elusive. Here, we combine genomic,single-cell imaging and biochemical approaches to chart themolecular pathways mediated by a key developmental regu-lator—Brachyury. Our results shed light on mechanistic insightsinto the ultrafine organization of Brachyury-bound enhancersand link Brachyury function to cellular differentiation andhousekeeping processes critical for coordinating early mouseembryogenesis.

Author contributions: M.L., P.D.T.V., R.T., and Z.L. designed research; M.L. and Z.L. per-formed research; M.L. and Z.L. analyzed data; and M.L., R.T., and Z.L. wrote the paper.

The authors declare no conflict of interest.

Freely available online through the PNAS open access option.

Data deposition: The data reported in this paper have been deposited in the Gene Ex-pression Omnibus (GEO) database, www.ncbi.nlm.nih.gov/geo (accession no. GSE54985).1To whom correspondence may be addressed. E-mail: liuz11@janelia.hhmi.org or tjianr@hhmi.org.

This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.1073/pnas.1402612111/-/DCSupplemental.

4478–4483 | PNAS | March 25, 2014 | vol. 111 | no. 12 www.pnas.org/cgi/doi/10.1073/pnas.1402612111

Dow

nloa

ded

by g

uest

on

Mar

ch 1

, 202

0

characterize gene expression changes upon differentiation, weperformed mRNA sequencing (mRNA-seq) with differentiatedembryoid bodies (EBs; EB day 4) and then compared gene ex-pression levels to ES cell data. We detected 2,464 up-regulatedand 1,329 down-regulated genes (statistics in Dataset S1). Con-sistent with a successful epiblast transition and primitive streakinduction, ES cell ground-state specific genes (Tbx3, Stra8, Pecam1,Esrrb, Gbx2, and Klf4) were dramatically down-regulated. However,mesoderm and definitive endoderm (DE) markers (Cer1, Eomes,Wnt3, Lhx1, Fgf8, Cxcr4, Foxa2, Sox17, and Gsc) (15) were stronglyup-regulated (SI Appendix, Figs. S1C and S2B). We also confirmedthe expression of Brachyury, Foxa2, and Sox17 at the protein levelby immunofluorescence staining (SI Appendix, Fig. S1D).

Brachyury Selectively Binds Key Developmental Genes. We appliedChIP sequencing (ChIP-seq) to our in vitro primitive streakformation system. To ensure specific detection of Brachyurybinding, we used two independent anti-Brachyury antibodies forChIP-seq experiments. By only calling peaks detected by bothantibodies, we identified 3,160 high-confidence Brachyury-boundregions in the mouse genome (Dataset S2). These sites displaygenome-wide biased association with transcription start sites(TSS; 61% are within 5 kb of a TSS vs. 9% of random controlregions) (Fig. 1 A and B). Intriguingly, we also observed a strongbiased localization of Brachyury bound at regions immediatedownstream of the TSS, particularly the 5′ UTR (Fig. 1C and SIAppendix, Fig. S1 E and F). As expected, many key developmentalgenes (Fgf8, Foxa2, Foxj1, Sox17, and Dusp6) were directly tar-geted by Brachyury (Fig. 1A and SI Appendix, Fig. S8). Wecorrelated Brachyury-binding site information with annotatedgenes genome-wide (<10 kb from TSS) and identified 1,942 genesbound by Brachyury (Dataset S3). To calculate overrepresentedGene Ontology terms associated with Brachyury binding, wenext performed Genomic Regions Enrichment of AnnotationsTool (GREAT) analysis (16) (Fig. 1D). We found that Brachyury-bound regions were significantly enriched around genes involvedin the formation of primary germ layers (P < 1.4E-9), mesodermcell migration (P < 3.0E-7), and mesoderm formation (P < 1.2E-5).Recently, Brachyury was found to express in DE progenitors, andBrachyury-positive cells were able to efficiently differentiateinto definitive endodermal cell types in vitro and in vivo (8, 9).

In line with these findings, Brachyury-bound regions were alsoenriched around genes involved in endoderm development(P < 3.9E-09) and formation (P < 2.6E-05).

Evolutionary Conservation of Brachyury-Mediated Regulation. Withrecent Brachyury homolog Ntl ChIP-on-chip (zebrafish) and XBraChIP-seq (Xenopus tropicalis) data available (12, 13), we nextasked whether a similar set of target genes identified in thesestudies also bound Brachyury in the mouse. Intriguingly, wefound that ∼10∼13% of the Ntl (zebrafish) and Xbra (X.tropicalis)target genes (zebrafish, 28/218; X. tropicalis, 113/1,040) have acorresponding homologous gene bound by Brachyury in mouse(Fig. 1E and Dataset S3). For example, nine genes (Dusp6,Mesp2, Hoxd8, Cdx2, Irx3, Lefty1, Msgn1, Dhrs3, and Sp5) werebound by Brachyury in all three species (Fig. 1F). We also ob-served Brachyury-bound genes such as Fgf8, Sox17, Foxa2, Tbx3,Wnt9b, Wnt5b, and Rest that were common targets in mouse andX. tropicalis but not in zebrafish (Fig. 1F and Dataset S3). Con-versely, Igf1r, Brf1, Gata2, Zfand5, Sox11, Mycn, Notch3, etc. arecommon targets in mouse and zebrafish but not in X. tropicalis. It isworth noting that the work performed in zebrafish used the pro-moter-restricted ChIP-on-chip technique (12 kb around the TSSof ∼11,000 genes) to map what is likely to be a relatively in-complete set of Brachyury-binding sites. We would thereforeexpect to observe larger function overlaps between mouse andzebrafish targets when more sensitive binding-site mappingtechniques were used.

Map High-Resolution Brachyury-Binding Sites by ChIP-Exonuclease.Previous in vitro studies suggested that Brachyury selectivelybinds a 7-bp DNA motif (TCACACCT) (17–19). However, sur-prisingly, this motif was not evident from Brachyury ChIP-on-chipdata in a recent report (20) and in our ChIP-seq datasets. Toresolve this discrepancy, we applied the much-higher-resolutionChIP-exonuclease (ChIP-exo) technique to our in vitro primitivestreak formation system. Exonuclease treatment steps in the ChIP-exo protocol remove extraneous flanking DNA from the cross-linked protein–DNA complexes (21), allowing a more precise de-lineation of DNA sequences recognized and bound by Brachyury(SI Appendix, Fig. S3A). Indeed, after peak calling by GeneTrack(22), we were able to detect 11,503 bifurcated ChIP-exo peaks

Fig. 1. Brachyury targets the core promoters andproximal enhancers (PEs) of key developmental genes.(A) Representative tracks for Brachyury-bound seq-regions associated with key developmental genes (theannotated gene length as indicated by the arrow:Upper Left, Fgf8, 6.087 kb; Upper Right, Foxa2,3.028 kb; Lower Left, Foxj1, 4.561kb; and Lower Right,Dusp6, 4.259 kb). y-axis limit is from 3 to 50 reads forchip-seq data. Auto-scale was used for displayingmRNA-seq tracks. Two independent antibodies (Ab1and Ab2) against Brachyury were used. PreimmuneIgG served as a negative control. mRNA-seq data fromcontrol EBs at day 4 are also displayed. (B) AbsoluteTSS proximity fraction curve of Brachyury-boundregions (blue) and that of randomly selected genomicregions (gray). (C) TSS proximity fraction curve ofBrachyury-bound regions (green). Brachyury-boundregions display a biased localization downstream of theTSS. (D) Enriched Annotation terms associated withSeq-regions (see Dataset S2 for more information;calculated by GREAT). (E) Venn diagram of evolutionconservation analyses of common Brachyury targetgenes in mouse, zebrafish, and X. tropicalis. (F) Listsof selected common Brachyury target genes in allthree species (M&Z&X) or in two species (M&Z;M&X). M, mouse; X, X. tropicalis; Z, zebrafish. See alsoDatasets S2 and S3.

Lolas et al. PNAS | March 25, 2014 | vol. 111 | no. 12 | 4479

DEV

ELOPM

ENTA

LBIOLO

GY

Dow

nloa

ded

by g

uest

on

Mar

ch 1

, 202

0

[11,000 (11k) dataset]. The average size of the binding sites was∼11 base pairs (bp), compared with 823 bp for a typical ChIP-seqbinding region. This significant improvement in resolution (∼76-fold) provided us with a more accurate view of Brachyury–enhancer DNA interactions (SI Appendix, Fig. S3B). ChIP-exoallowed us to successfully identify an overrepresented DNAmotif (TCACA) (1k dataset, top 10% of 11k dataset) that matchedwell with the Brachyury consensus sequence (P < 2.22E-16) (SIAppendix, Fig. S3 C and D). Genome-wide scanning (11k data-set) further confirmed that 37% of the Brachyury exo-peaks (4,299of 11,503) contain this motif (Fig. 2D). As expected, DNA probescontaining these 5 bp were sufficient to interact specificallywith Brachyury in EMSAs (SI Appendix, Fig. S4 A and B). Super-shift assays confirmed that this binding is Brachyury-specific (SIAppendix, Fig. S4C). Strikingly, we observed that there are fre-quently multiple Brachyury-binding sites within each ChIP-seq–defined enhancer region. For example, Foxa2, Cer1, and Wnt8bproximal enhancers (PEs) contain as many as eight Brachyuryexo-peaks (SI Appendix, Fig. S3E). On average, there are ∼2.4Brachyury-binding sites within each ChIP-seq region (Fig. 2A).

Unique Sequence Signatures in Brachyury-Bound Enhancers. An im-portant finding of the ENCODE (Encyclopedia of DNA Ele-ments) project was that many eukaryotic genes are regulated bymultiple TFs cobinding to neighboring sites in the genome (23).To test whether this characteristic is also the case for Brachyury-bound enhancers, we performed a motif search within 50 bparound Brachyury ChIP-exo peaks (1k dataset). We recoveredfive overrepresented motifs (Fig. 2B and SI Appendix, Fig. S5A).Importantly, four of the five identified motifs were confidentlymatched to known TF binding sequences (Brachyury, Myf fam-ily, E2F1, and Sox17) (Fig. 2C). In agreement with bona fide TFbinding events, a positional preference plot confirmed the en-richment for these motifs only at Brachyury exo-peaks but notwithin their flanking extended genomic regions (±200 bp) (Fig.2E). Consistent with the notion of combinatorial regulation, we

observed extensive double- and triple-motif colocalization (11kdataset) (Fig. 2D and SI Appendix, Fig. S5B), and the frequenciesof colocalization were significantly higher (up to sixfold) thanvalues calculated for random control genomic regions (Fig. 2D).These sequence signatures identified here provided us with valu-able information to further dissect complex molecular trans-actions at Brachyury-bound enhancers.

Brachyury Functions as an Activator. To probe how Brachyuryregulates its target genes, we performed mRNA-seq to analyzegene-expression changes after Brachyury depletion by two in-dependent lentivirus-mediated shRNAs (SI Appendix, Fig. S6A).Compared with nontarget shRNA control, we detected 567 down-regulated and 806 up-regulated genes common to the two distinctBrachyury shRNA-mediated knockdowns. Next, we set out to testwhether there is a biased association of Brachyury sites with par-ticular gene categories (up, down, and unchanged) upon Brachyurydepletion. To do so, we first assigned Brachyury-bound regionsthat were within 5 kb of the TSS to genes genome-wide. Then, weanalyzed the association between Brachyury-bound regions witheach gene category (SI Appendix, Fig. S6B). Notably, we only de-tected a biased association of Brachyury sites with genes that weredown-regulated upon loss of Brachyury, whereas the oppositecorrelation was seen for up-regulated genes. This genome-widebiased association strongly favors the notion that, upon induction,Brachyury binds to the promoter and enhancer regions of targetgenes and activates transcription. Consistent with this model, rep-resentative Brachyury-bound regions and, likewise, a synthetic DNAelement containing four repeats of the Brachyury DNA-bindingmotif efficiently elevated the gene expression of a firefly lucif-erase reporter in a Brachyury-dependent manner in human 293Tcells (SI Appendix, Figs. S6C and S7).

Brachyury Coordinates Gene Expression During in Vitro StreakFormation. To dissect how Brachyury regulates gene expressionat the single-cell level, we focused on two Brachyury direct target

Fig. 2. Unique sequence signatures at Brachyury-bound enhancers. (A, Left) The distribution of ChIP-exo peaks per ChIP-seq region. (Right) The modularbinding of Brachyury at single enhancers. (B, Left)Raw sequencing reads (forward, blue; reverse, green)of 200-bp genomic regions containing DNA motifs 2,3, and 4 that were de novo recovered by the gimme-motifs program (40) using the 1k dataset, centered atthe motif midpoint. (Right) Color chart representa-tion of 20 bp of sequence located around the mid-point of motif 1 ordered as in Left. (C) Motifs 1, 2, 3,and 4 were confidently matched to the known TFbinding sequences (T, Myf, E2F1, and Sox17) by Jaspar.(D, Left) Number of 50-bp genomic regions aroundexo-peaks (11k dataset) that contain either single ordouble motifs. (Right) Fold enrichment when com-paring numbers in Left with numbers that were cal-culated with randomized control genomic regions (SIAppendix, SI Methods). (E, Left) Probability-densitydistribution of motifs 1–5 across 400-bp genomic re-gions surrounding exo-peaks (centered by the peakmidpoint). (Right) Schematic diagram illustratingcombinatorial regulation by colocalization of multi-ple transfactors at Brachyury-bound enhancers.

4480 | www.pnas.org/cgi/doi/10.1073/pnas.1402612111 Lolas et al.

Dow

nloa

ded

by g

uest

on

Mar

ch 1

, 202

0

genes, Foxa2 and Sox17 (Fig. 1A and SI Appendix, Fig. S8B).Interestingly, the target gene analysis comparing divergent speciesindicated that both Foxa2 and Sox17 were also targeted by theBrachyury homolog (XBra), suggesting that the regulatory linkbetween Brachyury, Foxa2, and Sox17 might be evolutionarilyconserved (Fig. 1 E and F). Furthermore, Brachyury depletiondecreased Foxa2 and Sox17 expression at both protein and mRNAlevels (Fig. 3A and SI Appendix, Fig. S9 A and B). After doubleor triple immunofluorescence staining, we performed single-cellanalysis to test the correlation of expression between genes (Fig. 3Band SI Appendix, Fig. S9 C and D). Collectively, our single-cellimaging analysis revealed several important correlations regardingthe relationship between Brachyury, Foxa2, and Sox17 (Fig. 3Band SI Appendix, Fig. S9C). Firstly, Foxa2high and Sox17high cellswere mutually exclusive, as highlighted by the negative correlationbetween their expression levels. We also identified a negative cor-relation between Sox17 and Brachyury. In contrast, the expres-sion of Foxa2 was always positively correlated with Brachyuryexpression (Fig. 3B). We also observed a population of cells withhigh levels of Brachyury and medium-to-low levels of Sox17 andFoxa2 (SI Appendix, Fig. S9D), suggesting coexpression ofBrachyury, Foxa2, and Sox17 within the same individual cells.Together with genomic data, our single-cell imaging results

favor a model in which, as differentiation occurs, Brachyury firsttargets and promotes the expression of Foxa2 and Sox17. How-ever, high levels of Sox17 would, in turn, directly or indirectlyrepress the expression of Brachyury (Fig. 3H). This potentialnegative feedback may explain the Sox17highBrachyurylowFoxa2low

cell population. To directly test this model, we overexpressed Sox17after 2 days of differentiation when cells are at the epiblast stage.Indeed, we observed significant down-regulation of Brachyuryat EB day 4 (Fig. 3C).

To further test this model in vivo, we first asked whetherBrachyury, Sox17, and Foxa2 also displayed gene-expressionpatterns that spatially overlap during early mouse development.We first compared publicly available in situ hybridization (ISH)gene expression data (24). Indeed, we detected coexpression ofBrachyury and Foxa2 at node/DE regions and coexpression ofFoxa2 and Sox17 at DE regions (SI Appendix, Fig. S9E). Consis-tent with our in vitro results, we found that Brachyury and Sox17expression was negatively correlated in the definite endodermregion. If the model derived from our in vitro experiments iscorrect, we would expect to detect a BrachyurymediumSox17medium

cell population, suggesting an intermediate transition statewherein Brachyury first activates the expression of Sox17 and in-creasing levels of Sox17 gradually represses the expression ofBrachyury. To further probe the coregulated gene expression atthe single-cell level, we performed confocal imaging experimentsof coimmunostained 7.0∼7.5 mouse embryos (Fig. 3 D and E, SIAppendix, Fig. S10, and Movies S1–S4). As expected, we detecteda BrachyurymediumSox17medium cell population in the DE region,but not at other regions, in embryonic day 7.5 (E7.5) embryos(Fig. 3 D and F, SI Appendix, Fig. S10A, and Movie S1). Inter-estingly, BrachyurymediumSox17medium cells are generally foundlocated more distal to the midline of the notochord-primitivestreak compared with BrachyuryhighSox17low cells, suggestingthat this population of cells might have a distinct morphogeneticmode during early embryogenesis (Fig. 3F Right). Consistentwith the ISH results, we also detected coexpression of Brachyuryand Foxa2 within the same individual cells within node and DEregions (Fig. 3E, SI Appendix, Fig. S10 B and C, and Movies S2–S4). Brachyury-positive cells express Foxa2 at significantly higherlevels than Brachyury-negative cells (Fig. 3G). This finding is ingood agreement with our in vitro results showing that Brachyurydirectly targets and activates the transcription of Foxa2 and that

Fig. 3. A gene-regulatory system of Brachyury,Foxa2, and Sox17. (A) Western blot analysis ofFoxa2 and Sox17 protein levels after Brachyuryknockdown. ES cells and EB day 4 without Activin-Ainduction served as negative controls. (B, Left) Sin-gle-cell gene-expression correlation analysis. Seg-mented cell areas in the Sox17 channel weredetermined by proper computational steps (binarycontrast enhancing and particle size/circularity fil-tering). Then, the mean fluorescent intensities fromthe two channels were plotted and tested for cor-relation. (Right) Negative correlation between theprotein levels of Sox17 and Foxa2 in single cells.There is a negative correlation between the proteinlevels of Sox17 and Brachyury when segmented cellareas were determined by the Sox17 channel. Thereis a positive correlation between the protein levelsof Brachyury and Foxa2 in single cells when seg-mented cell areas were determined by the Foxa2channel. A.U., arbitrary units. (C) Sox17 repressesthe expression of Brachyury during in vitro streakformation. GFP served as transfection control. EBs(day 2; SI Appendix, Fig. S1A) were first dissociatedby trypsin into single cells, and then electroporationwas performed with either Sox17 or GFP over-expressing plasmid. Cells were reaggregated, andWestern blot analysis of Brachyury, Sox17, and GFP was performed at EB day 4. (D) Whole-mount Brachyury and Sox17 Immunostaining confocal analysis ofE7.5 embryos. DE region was selected to show. For other regions, see Movie S1. Cells with medium expression levels of Brachyury and Sox17 are indicated bygray arrows. (Magnification: Upper, 10×; Lower, 40× oil; scale bars: Upper, 100 μm; Lower, 25 μm.) (E) Whole-mount Brachyury and Foxa2 Immunostainingconfocal analysis of E7.5 embryos. Cross-section covering both primitive streak and DE was selected to show. For other regions, see Movie S2. (Magnification:Upper, 10×; Lower, 40× oil; scale bars: Upper, 100 μm; Lower, 25 μm.) (F, Left) Three populations of cells were observed in Brachyury and Sox17 embryoImmunostaining analysis. High expression of Brachyury but low expression of Sox17 (B++/S−) is shown. Medium expression of Brachyury and medium ex-pression of Sox17 (B+/S+), as indicated by arrows in D. is shown Low expression of Brachyury but high expression of Sox17 (B−/S++) is shown. (Right) B+/S+cells are generally further away from the primitive streak midline than B++/S− cells. (G) At the node/DE region, Brachyury-positive cells generally have higherFoxa2 levels than Brachyury-negative cells (also see SI Appendix, Fig. S10 B and C and Movies S2–S4). (H) A diagram illustrating the gene-regulatory rela-tionships between Brachyury, Foxa2, and Sox17.

Lolas et al. PNAS | March 25, 2014 | vol. 111 | no. 12 | 4481

DEV

ELOPM

ENTA

LBIOLO

GY

Dow

nloa

ded

by g

uest

on

Mar

ch 1

, 202

0

BrachyuryhighFoxa2high and BrachyurylowSox17high cells are mu-tually exclusive. Collectively, these results support the modelderived from in vitro experiments and suggest that that theregulation of Foxa2 and Sox17 by Brachyury might be region-specific. Specifically, we note that Sox17 and Foxa2 are also ex-pressed in some embryonic tissues, such as the primitive endo-derm, where Brachyury is largely absent (Fig. 3 D and E andMovies S1–S4) (25). Thus, it is likely that these two genes canbe regulated by distinct factors and mechanisms at different de-velopmental stages. It appears that, at least during streak forma-tion, the expression of both genes displays significant reliance onBrachyury (Fig. 3A and SI Appendix, Fig. S9 A and B).

Probing in Vivo Brachyury Function. To identify Brachyury targetgenes in vivo and elucidate how Brachyury-mediated regulationcontributes to early mouse developmental homeostasis and co-ordination, we made use of the reported Brachyury mutant Tc(Curtailed) mouse model (26). Importantly, a single copy of theWT Brachyury transgene rescues the Tc mutant phenotype (27).Thus, the Tc mouse line should be a useful model for studyingBrachyury loss of function. As reported, we found that, by E8.0,Tc/Tc embryos become severely shortened compared with WTembryos (SI Appendix, Fig. S11A), and by E10.0, Tc/Tc embryosshowed dramatic axial and neural fold closure defects (SI Ap-pendix, Fig. S11B). We next performed mRNA-seq to comparegene-expression profiles of WT and Tc/Tc embryos at both E7.5∼8.0 and E10.0 ∼10.5 (SI Appendix, Fig. S11C). The Brachyurygene becomes highly expressed at E7.5 ∼8.0 (Fig. 4A), suggestingthat these are reasonable stages at which to study the directimpact of Brachyury. Next, we correlated Brachyury target geneinformation with up-/down-regulated genes in E7.5 ∼8.0 Tc/Tcembryos (Fig. 4B and Dataset S3). We detected 70 down-regu-lated and 53 up-regulated functional Brachyury target genes.Thus, it is likely that, although Brachyury targets a large number(1,942) of genes in the genome, only a small fraction of targetedgenes showed gene-activation dependence on Brachyury, con-sistent with the previous observation that only 81 genes of ∼1,400targeted genes are dependent of Xbra in Xenopus (12).

Brachyury-Mediated Regulation in Early Development. We next ex-amined whether our integrated genome-wide data are consistentwith previous genetic studies and whether our data could link themolecular function of Brachyury to unique developmental andcellular processes. For this analysis, we only considered genesthat are directly targeted and up-regulated by Brachyury in vitroand/or in vivo. We also scored each target gene according to itsevolutionary conservation (Fig. 4C and Dataset S4). Indeed,consistent with the positive feedback loop between T and Fgf8(28–30), we found that Fgf8 is under the control of four Brachyury-bound enhancer regions (Fig. 1A) and that Fgf8 is down-regulatedin both in vitro and in vivo perturbation experiments (Dataset S4).Fgf8 is also a common Brachyury target in mouse and X. tropicalis(Fig. 1F). In parallel, we found that Brachyury targets the Dusp6gene, a negative regulator of Fgf signaling (31) (Fig. 1A). Impor-tantly, this regulatory link between Brachyury and Dusp6 was ob-served in all three species (Fig. 1F), suggesting that this pathway ishighly conserved. Thus, it is likely that, after first potentiating theFgf8 signaling, Brachyury also initiates a negative feedback controlto ensure the proper attenuation of signaling. Previous geneticstudies showed that both Foxa2-null and Brachyury-mutant micedisplayed node/notochord defects that manifest between E8 andE10 (27, 32). Here, we found that Foxa2 is a functional and directBrachyury target gene (Figs. 1A and 3A and SI Appendix, Fig. S9 Aand B), suggesting that loss of Foxa2 expression in Brachyury-positive cells might partially explain some of the Brachyury mutantphenotypes. These results are also consistent with the proposedregulation between Brachyury and Foxa2 in organizer derivatives(node/notochord) described by Tamplin et al. (33).

Fig. 4. Brachyury-mediated gene-regulatory network. (A) ISH assay map-ping Brachyury gene expression in E7.75 WT embryo. PS, primitive streak.The publicly shared data were obtained from the EMAGE gene expressiondatabase under the citation agreement (EMAGE ID 104; ref. 24). (B) Tc/Tcembryo differential gene-expression data (Dataset S1) are correlated withBrachyury-target gene information (Dataset S3) to identify functional in vivoBrachyury targets. A total of 70 up-regulated (P = 1.14E-6) and 53 down-regulated (P = 0.036) Brachyury target genes were identified (listed inDataset S4). (C) Functional Brachyury target genes and their known func-tions. Selected genes directly targeted and activated by Brachyury are listed.A confidence score is assigned to each gene by weighting in vitro and in vivoperturbation and evolution conservation results. See Dataset S4 for calcu-lation details. A value of 4 denotes the highest confidence. (D) Differentialgene-expression analysis to characterize downstream developmental con-sequences of Brachyury loss-of-function with mRNA-seq to identify up-/down-regulated genes in Tc/Tc embryos at E10.0∼10.5. The log2 scale values offragments per kilobase per million (FKPM) for genes in both WT and Tc/Tcsamples are plotted in the graph. Dark green, twofold down; light green,eightfold down; orange, twofold up; red, eightfold up. The hematopoiesis andskeletal muscle markers (Hba-x, Hbb-y, and Myog) are highlighted in red circles.See also SI Appendix, Fig. S11 and Datasets S1 and S4.

4482 | www.pnas.org/cgi/doi/10.1073/pnas.1402612111 Lolas et al.

Dow

nloa

ded

by g

uest

on

Mar

ch 1

, 202

0

We identified the Mab21l2 gene (SI Appendix, Fig. S8A), apotential bone morphogenetic protein (BMP) antagonist (34), asa Brachyury target gene. Although the regulation of Mab21l2 byBrachyury is only found in the mouse, theMab21l2 gene is down-regulated in both in vitro and in vivo Brachyury loss-of-functionexperiments (Dataset S4). Our results further suggest thatBrachyury might coordinate primary cilia formation and left–right axis patterning via regulation of Foxj1, Cdh2, and Tbx6—all of which have been shown to be critical for the establishmentof early embryo left–right asymmetry (35–37). Consistent withthe function of Brachyury in regulating morphogenetic cell move-ments, Brachyury up-regulates the expression of key epithelial–mesenchymal transition (EMT) and cell-mobility regulators (Sox4and Bves). The functional link between Brachyury and Sox4 alsooffers mechanistic insights into recent observations that Brachyurypromotes the EMT and metastatic behavior of human tumor cells(38). We also identified other cancer-related genes (Tpd52l1, Ybx1,and Irf1) as direct targets of Brachyury.Strikingly, we found that Brachyury directly activates a set of

hematopoietic regulators (Mixl1, Mllt3, NKX2-3, Sox6, POU4F1,and CD93). Most of these regulatory pathways are not conservedin zebrafish and Xenopus, suggesting that these are evolutionarilydivergent networks (Dataset S4). Tc/Tc embryo mRNA-seq re-sults at later stages of development (E10.0∼10.5) confirmed down-stream defects in hematopoiesis. Specifically, the fetal hemoglobin

genes (Hba-x and Hbb-y) were significantly down-regulated (Fig.4D). Importantly, these results agree well with previous observa-tions that hemangioblast commitment is initiated by Brachyury-positive cells in the primitive streak of the mouse embryo (39) andthus further establish a molecular link between Brachyury and he-matopoietic lineage specification.Consistent with the role of Brachyury in directing mesoderm

lineages, Brachyury targets and promotes the expression of muscleand chondrocyte regulators (Msgn1, Pax7, My17, Mfhx4, andSox9). As expected, skeletal muscle development (Myog) wascompromised in the E10.0∼10.5 Tc/Tc embryos (Fig. 4D), con-sistent with its reported regulatory function in zebrafish (13).

MethodsMouse D3 (ATCC) ES cells were used for in vitro differentiation experiments. TheTc/J mouse line was obtained from the Jackson Laboratory. Custom-made Bra-chyury antibodies were used for ChIP-seq/exo experiments. Deep-sequencingwas performed with the illumina HiSeq system. Detailed methods are availablein SI Appendix, SI Methods. Primer information is available in Dataset S5.

ACKNOWLEDGMENTS. We thank M. Haggart, S. Zheng, C. Morkunas,S. Moorehead, and Katie McDole for assistance. M.L. was a visiting PhDstudent at Janelia Farm Research Campus and was also supported partiallyby Conicyt Program PFB16. Z.L. is a Janelia junior fellow and was also supportedas a predoctoral California Institute for Regenerative Medicine fellow atUniversity of California, Berkeley.

1. Showell C, Binder O, Conlon FL (2004) T-box genes in early embryogenesis. Dev Dyn229(1):201–218.

2. Chesley P (1935) Development of the short-tailed mutant in the house mouse. J ExpZool 70(3):429–459.

3. Yanagisawa KO (1990) Does the T gene determine the anteroposterior axis ofa mouse embryo? Jpn J Genet 65(5):287–297.

4. Gruneberg H (1958) Genetical studies on the skeleton of the mouse. XXIII. The de-velopment of brachyury and anury. J Embryol Exp Morphol 6(3):424–443.

5. Gluecksohn-Schoenheimer S (1938) The development of two tailless mutants in thehouse mouse. Genetics 23(6):573–584.

6. Herrmann BG, Kispert A (1994) The T genes in embryogenesis. Trends Genet 10(8):280–286.

7. King T, Beddington RS, Brown NA (1998) The role of the brachyury gene in heartdevelopment and left-right specification in the mouse. Mech Dev 79(1-2):29–37.

8. Kubo A, et al. (2004) Development of definitive endoderm from embryonic stem cellsin culture. Development 131(7):1651–1662.

9. Gouon-Evans V, et al. (2006) BMP-4 is required for hepatic specification of mouseembryonic stem cell-derived definitive endoderm. Nat Biotechnol 24(11):1402–1411.

10. Kita-Matsuo H, et al. (2009) Lentiviral vectors and protocols for creation of stablehESC lines for fluorescent tracking and drug resistance selection of cardiomyocytes.PLoS ONE 4(4):e5046.

11. Liu Z, Scannell DR, Eisen MB, Tjian R (2011) Control of embryonic stem cell lineagecommitment by core promoter factor, TAF3. Cell 146(5):720–731.

12. Gentsch GE, et al. (2013) In vivo T-box transcription factor profiling reveals jointregulation of embryonic neuromesodermal bipotency. Cell Rep 4(6):1185–1196.

13. Morley RH, et al. (2009) A gene regulatory network directed by zebrafish No tailaccounts for its roles in mesoderm formation. Proc Natl Acad Sci USA 106(10):3829–3834.

14. Gadue P, Huber TL, Paddison PJ, Keller GM (2006) Wnt and TGF-beta signaling arerequired for the induction of an in vitro model of primitive streak formation usingembryonic stem cells. Proc Natl Acad Sci USA 103(45):16806–16811.

15. Tesar PJ, et al. (2007) New cell lines from mouse epiblast share defining features withhuman embryonic stem cells. Nature 448(7150):196–199.

16. McLean CY, et al. (2010) GREAT improves functional interpretation of cis-regulatoryregions. Nat Biotechnol 28(5):495–501.

17. Naiche LA, Harrelson Z, Kelly RG, Papaioannou VE (2005) T-box genes in vertebratedevelopment. Annu Rev Genet 39:219–239.

18. Kispert A, Hermann BG (1993) The Brachyury gene encodes a novel DNA bindingprotein. EMBO J 12(12):4898–4899.

19. Bruneau BG, et al. (2001) A murine model of Holt-Oram syndrome defines roles of theT-box transcription factor Tbx5 in cardiogenesis and disease. Cell 106(6):709–721.

20. Evans AL, et al. (2012) Genomic targets of Brachyury (T) in differentiating mouseembryonic stem cells. PLoS ONE 7(3):e33346.

21. Rhee HS, Pugh BF (2011) Comprehensive genome-wide protein-DNA interactionsdetected at single-nucleotide resolution. Cell 147(6):1408–1419.

22. Albert I, Wachi S, Jiang C, Pugh BF (2008) GeneTrack—a genomic data processing andvisualization framework. Bioinformatics 24(10):1305–1306.

23. Bernstein BE, et al.; ENCODE Project Consortium (2012) An integrated encyclopedia ofDNA elements in the human genome. Nature 489(7414):57–74.

24. Richardson L, et al. (2010) EMAGE mouse embryo spatial gene expression database:2010 update. Nucleic Acids Res 38(Database issue):D703–D709.

25. Niakan KK, et al. (2010) Sox17 promotes differentiation in mouse embryonic stemcells by directly regulating extraembryonic gene expression and indirectly antago-nizing self-renewal. Genes Dev 24(3):312–326.

26. Searle AG (1966) Curtailed, a new dominant T-allele in the house mouse. Genet Res7(1):86–95.

27. Stott D, Kispert A, Herrmann BG (1993) Rescue of the tail defect of Brachyury mice.Genes Dev 7(2):197–203.

28. Ciruna B, Rossant J (2001) FGF signaling regulates mesoderm cell fate specificationand morphogenetic movement at the primitive streak. Dev Cell 1(1):37–49.

29. Tam PP, Loebel DA (2007) Gene function in mouse embryogenesis: Get set for gas-trulation. Nat Rev Genet 8(5):368–381.

30. Sun X, Meyers EN, Lewandoski M, Martin GR (1999) Targeted disruption of Fgf8causes failure of cell migration in the gastrulating mouse embryo. Genes Dev 13(14):1834–1846.

31. Ekerot M, et al. (2008) Negative-feedback regulation of FGF signalling by DUSP6/MKP-3 is driven by ERK1/2 and mediated by Ets factor binding to a conserved sitewithin the DUSP6/MKP-3 gene promoter. Biochem J 412(2):287–298.

32. Ang SL, Rossant J (1994) HNF-3 beta is essential for node and notochord formation inmouse development. Cell 78(4):561–574.

33. Tamplin OJ, et al. (2008) Microarray analysis of Foxa2 mutant mouse embryos revealsnovel gene expression and inductive roles for the gastrula organizer and its de-rivatives. BMC Genomics 9:511.

34. Baldessari D, Badaloni A, Longhi R, Zappavigna V, Consalez GG (2004) MAB21L2,a vertebrate member of the Male-abnormal 21 family, modulates BMP signaling andinteracts with SMAD1. BMC Cell Biol 5(1):48.

35. Tamakoshi T, et al. (2006) Roles of the Foxj1 and Inv genes in the left-right de-termination of internal organs in mice. Biochem Biophys Res Commun 339(3):932–938.

36. García-Castro MI, Vielmetter E, Bronner-Fraser M (2000) N-Cadherin, a cell adhesionmolecule involved in establishment of embryonic left-right asymmetry. Science288(5468):1047–1051.

37. Hadjantonakis AK, Pisano E, Papaioannou VE (2008) Tbx6 regulates left/right pat-terning in mouse embryos through effects on nodal cilia and perinodal signaling.PLoS ONE 3(6):e2511.

38. Fernando RI, et al. (2010) The T-box transcription factor Brachyury promotes epi-thelial-mesenchymal transition in human tumor cells. J Clin Invest 120(2):533–544.

39. Huber TL, Kouskoff V, Fehling HJ, Palis J, Keller G (2004) Haemangioblast commit-ment is initiated in the primitive streak of the mouse embryo. Nature 432(7017):625–630.

40. van Heeringen SJ, Veenstra GJ (2011) GimmeMotifs: A de novo motif predictionpipeline for ChIP-sequencing experiments. Bioinformatics 27(2):270–271.

Lolas et al. PNAS | March 25, 2014 | vol. 111 | no. 12 | 4483

DEV

ELOPM

ENTA

LBIOLO

GY

Dow

nloa

ded

by g

uest

on

Mar

ch 1

, 202

0