Cohesin-dependent regulation of Runx genes

Runx proteins form part of complexes called core binding factors (CBFs);multi-lineage transcriptional regulators with roles in both proliferation and differentiation. CBFs act as dimers that incorporate one of three distinct DNA-binding α subunits (Runx1, Runx2 or Runx3) plus a common non-DNA-binding CBFβ subunit. The Runx component specifies the biological activity of CBF; Runx1 is an essential regulator of hematopoiesis, Runx2 is involved in osteogenesis, and Runx3 is important in neurogenesis and gastric epithelial cell growth control (Blyth et al., 2005; Ito,2004). Deregulation of Runx function is commonly associated with disease. For example, Runx1 is involved in several chromosomal translocations underlying human leukemias. In addition, changes in the dose of Runx1can contribute to neoplasia (Blyth et al.,2005). An important route toward understanding the pathology of Runx-mediated cancers is the elucidation of factors that control either the expression of Runx genes, or the activity/stability of their protein products.

In early zebrafish embryogenesis, runx1 is expressed in two discrete hematopoietic regions; the anterior lateral-plate mesoderm (ALM),which generates primitive myeloid cells, and the posterior lateral-plate mesoderm (PLM) from which primitive erythroid cells develop(Hsia and Zon, 2005). By around 18 hours post-fertilization (h.p.f.), cells in the PLM have migrated medially to form a central rod of hematopoietic precursors: the intermediate cell mass (ICM). Hematopoietic expression of runx1 is downregulated in all but the most posterior cells of the ICM at 21 h.p.f., and subsequently reappears in definitive hematopoietic precursors in the ventral wall of the dorsal aorta by 24 h.p.f. (Kalev-Zylinska et al., 2002). runx1 is also expressed in Rohon-Beard(RB) mechanosensory neurons and in specific neuronal cells(Kalev-Zylinska et al.,2002).

Studies in zebrafish (Burns et al.,2005; Gering and Patient,2005) and mice (Nakagawa et al., 2006) have shown that Runx1 is a transcriptional target of Notch signaling during definitive hematopoiesis. In zebrafish,Hedgehog signaling is required for the migration of hematopoietic progenitors to the midline, and for the subsequent formation of runx1⁺definitive precursors (Gering and Patient,2005). Early PLM runx1 expression appears to be downstream of a Hox pathway regulated by caudal-related homeobox genes cdx1a and cdx4(Davidson et al., 2003; Davidson and Zon, 2006). In addition the timely initiation (but not maintenance) of runx1expression depends on the transcription factors Scl and Lmo2(Patterson et al., 2007; Patterson et al., 2005). Other factors known to contribute to Runx gene regulation include the BMP signaling pathway (Pimanda et al.,2007), and epigenetic modifications, such as promoter methylation(Lau et al., 2006; Mueller et al., 2007).

To search for potential regulators of runx1 expression in zebrafish, we conducted an in situ hybridization-based haploid genetic screen of F1 females carrying mutations generated by ethylnitrosourea (ENU). We isolated a mutant, termed nz171, which lacks some neuronal, and all hematopoietic runx1 expression in early embryogenesis. Through a positional cloning and candidate gene approach, we determined that the gene underlying nz171 was rad21, an integral subunit of mitotic cohesin.

Cohesin is a protein complex composed of four major subunits: SMC1, SMC3,RAD21 and SA1 (or SA2), which interact to form a giant ring-like structure. Mitotic cohesin acts as a `molecular glue' to hold replicated sister chromatids together until the onset of anaphase(Losada and Hirano, 2005; Nasmyth and Haering, 2005). Cohesin also has a DNA repair function(Watrin and Peters, 2006). Intriguingly, it seems that cohesin has additional non cell cycle-related functions (Dorsett, 2007; Hagstrom and Meyer, 2003). In Drosophila, loss of Nipped-B, a protein that loads cohesin onto chromosomes, affects expression of the cut gene(Dorsett et al., 2005; Rollins et al., 2004). The human ortholog of Nipped-B/Scc2 is NIPBL, and mutations in the NIPBLgene, or in the cohesin subunit SMC1, underlie the dominant developmental syndrome, Cornelia de Lange (CdLS)(Dorsett, 2007; Musio et al., 2006; Strachan, 2005). Although cohesin clearly has a role in development as well as in the cell cycle, the mechanisms underlying its developmental function are unknown.

Our study provides the first evidence of cohesin-dependent gene regulation in a vertebrate. We found that Runx gene expression depends on the function of the whole cohesin complex and that, as expected, loss of zebrafish cohesin subunits interferes with sister chromatid cohesion during mitosis. Differentiated blood cells were deficient in nz171 mutants, and we found that some of these cells could be rescued by microinjection of runx1. nz171 mutants also had neuronal defects and rad21 was robustly expressed in specific regions of the brain at 48 h.p.f., which might indicate a role in neuronal development. We observed a potent dosage effect of Rad21 on downstream gene expression, and determined that halving the dose of the rad21 gene reduces the expression of runx1 and proneural genes ascl1a and ascl1b. Our findings highlight a new role for cohesin in gene expression and development.

Zebrafish were maintained as described previously(Westerfield, 1995). AB strain males were mutagenized with four treatments of 2.5 mM ENU as described previously (Pelegri, 2002),and were bred with AB females to obtain F1 fish. Eggs from F1 females were fertilized in vitro with UV-irradiated sperm to produce haploid progeny(Pelegri, 2002), which were subjected to in situ hybridization with a runx1 riboprobe. The nz171 F1 female carrier was crossed to the WIK strain for meiotic mapping. Chromosome localization of the nz171 mutation was performed as described previously (Geisler, 2002). `Z marker' PCR primer sequences were obtained from the WashU Zebrafish Genome Resources Project(http://zfish.wustl.edu/)and subsequent mapping was performed as described by Geisler(Geisler, 2002).

The genomic sequence of zebrafish rad21 is available in Ensembl(gene ID ENSDARG00000006092). For mutation analysis, rad21 was amplified in overlapping segments from cDNA made from 100 pooled nz171 homozygous mutant embryos and 100 wild-type AB embryos (48 h.p.f.). Primers used were Forward-1, 5′-TTCAATGAGAGGAAACGGTTGC-3′; Reverse-1,5′-ATGATGTCACTGTAGGGCTCGG-3′; Forward-2,5′-TTCAATGAGAGGAAACGGTTGC-3′; Reverse-2,5′-GGAAGAGGCTGGTCAAAATCG-3′; Forward-3,5′-CCCACTTCGTCCTCAGTAAACG-3′; Reverse-3,5′-TGGTCTTGCTGTCCAACTCCTTC-3′; Forward-4,5′-AGACAACCGTGAGGCAGCATAC-3′; Reverse-4,5′-AGTGGAGTCAAGCAGCGAGTAAAC-3′; Forward-5,5′-ATGTGGAAGGAGACTGGAGGTGTG-3′; Reverse-5 5′-TACAATGTGGAAGCGTGGTCCC-3′; Forward-6,5′-TCAGGACCAAGAGGAGAGAAGGTG-3′; Reverse-6,5′-CAACAAGTAGTGAAACTGCGGAGTC-3′. Amplified fragments were cloned and sequenced, and the coding region mutation at nucleotide 829 was represented by two independent overlapping clones and fourfold sequence coverage.

In situ hybridization was performed as described previously(Kalev-Zylinska et al., 2002). Embryos to be sectioned were dehydrated in an ethanol series, embedded in JB-4(Polysciences), cut to a thickness of 5 μm and stained with Giemsa(BDH).

Bromodeoxyuridine (BrdU) incorporation was performed on whole embryos as described previously (Shepard et al.,2004). Mouse monoclonal anti-BrdU antibody (Zymed) was detected using a Vectastain ABC Kit (Vector Laboratories). Phosphorylated histone H3 was detected as described previously(Shepard et al., 2004). TUNEL staining was performed using the ApopTag Kit from Chemicon as described previously (Shepard et al.,2004).

Full-length zebrafish and human rad21 clones (RZPD) were subcloned into pCS2⁺. Amplified full-length rad21^G277Xwas subcloned using a ZeroBluntTOPO Kit (Invitrogen), sequenced entirely to confirm the mutation, and subcloned into pCS2⁺. mRNAs for microinjection were generated using a mMessage mMachine Kit (Ambion), and 100 pg of each was injected into nz171 mutant and sibling embryos at the 1-cell stage. Full-length zebrafish runx1 in pCS2⁺(Kalev-Zylinska et al., 2002)was transcribed as above and 200 pg was injected as above. Morpholino oligonucleotides were obtained from GeneTools LLC and diluted in water. For microinjection, 2 nl of morpholino was injected into the yolk of wild-type embryos from the 1- to 4-cell stages. Morpholino oligonucleotides targeting rad21 were: rad21UTRMO,5′-CACTACACCTGGAAGAAAACAG-3′; rad21ATGMO 5′-TCCTGCTTCACCCGCATTTTGTAAC-3′ (start codon underlined); rad21Splx3MO 5′-GATACAATACCTGGGCGGAAAG-3′(targets 3′ donor of exon 3). Morpholino oligonucleotides targeting smc3 were smc3UTRMO,5′-GCACAAAACACTCCTCAGAAAC-3′; smc3ATGMO,5′-TGTACATGGCGGTTTATGC-3′ (start codon underlined); smc3Splx1MO, 5′-GTGAGTCGCATCTTACCTG-3′ (targets 3′donor of exon 1) and smc3Splx5MO,5′-TTTCTTACTGGAAGTCTGTTGTCAG-3′ (targets 3′ donor of exon 5). All morpholinos were effective over the range of 0.5-3.0 pmol injected.

For immunofluorescence, embryos were fixed and stained with anti-Rad21(Chemicon International) 1:100, anti-α-tubulin (Sigma-Aldrich) 1:500,and DAPI as described previously (Shepard et al., 2004). FITC- or TRITC-conjugated secondary antibodies(Sigma, 1:500) were used. Flat-mounted samples were imaged using a Leica TCS SP2 confocal microscope.

For immunoblotting, embryos were deyolked in Ringer's solution with EDTA and PMSF, and 20 μg total protein was loaded per lane. For embryos under 12 h.p.f., entire single, or pools of up to 10 embryos were lysed in loading buffer. Sample processing and immunoblotting was performed as described previously (Westerfield, 1995)using anti-Rad21 (Chemicon, 1:500), or anti-α-tubulin (Sigma, 1:2000). Horseradish peroxidase-linked secondary antibodies (Sigma, 1:2000) and enhanced chemiluminescence were used according to the manufacturer's instructions (ECL Plus, Amersham Biosciences, Inc.). Signals were analyzed using Fuji LAS-3000 imager and Fuji Image Gauge software. For quantitative PCR, total RNA from pools of 30-100 embryos was extracted using Trizol(Invitrogen), DNAse-treated, and used to synthesize random-primed cDNA(Invitrogen, SuperScript III). SYBR green PCR Master Mix (Applied Biosystems)was used to amplify cDNA, and relative start quantities were normalized toβ -actin and wnt5a expression. Samples were analyzed using an Applied Biosystems Sequence Detection System 7900HT. Primers for quantitative PCR were designed using the Primer Express program (Applied Biosystems). Sequences are: runx1 forward,5′-AGACGTCTCCATCCTGGTCGTA-3′, reverse,5′-CCGTCAGCTCTGGACAGTGTAA-3′; rad21 forward,5′-CAGCATACAATGCCATCACCTT-3′, reverse,5′-ATTCAGGGTGAACTGCTGTGCTA-3′; ascl1a forward,5′-GGGCTCATACGACCCTCTGA-3′, reverse,5′-TCCCAAGCGAGTGCTGATATTT-3′; ascl1b forward,5′-CCACATGGTTCGACAGATACGA-3′, reverse,5′-CAGCATGCAGCAAATCAAAGAC-3′; β-actin5′-CGAGCAGGAGATGGGAACC-3′, reverse,5′-CAACGGAAACGCTCATTGC-3′; wnt5a forward,5′-GTTCGGCCGCGTCATG-3′, reverse,5′-TCGACTCACAGCATTCACAACA-3′.

The nz171 mutant was isolated in a haploid in situ hybridization screen because of a marked lack of runx1 expression in the PLM in early embryogenesis, and lack of blood at 48 h.p.f. In nz171 diploids, runx1 expression was confirmed lost from the ALM and PLM, but was retained in a subset of RB neurons at the 14-somite stage(Fig. 1A,G compared with D). The runx3 gene is expressed in a subset of RB cells and in the trigeminal ganglia (TGG) during early embryogenesis(Kalev-Zylinska et al., 2003)(Fig. 1E). Strikingly, its expression was completely absent in nz171 mutants at the 14-somite stage(Fig. 1H). Although both runx1 and runx3 are downregulated in nz171, their common binding partner, cbfb was expressed normally in neuronal and hematopoietic tissues in equivalent embryos(Fig. 1I compared with 1F). This indicates that the cell types that would normally express runx1and runx3 are present, and that loss of runx1 and runx3 expression was not due to gross developmental defects. Early neurogenesis was abnormal in nz171 embryos; the TGG failed to develop properly(although they are specified), and the number of RB neurons was reduced by 30-40% (data not shown). Furthermore, although genes that mark blood vessel formation were expressed (see Fig. S1 in the supplementary material), no visible circulation developed. From the 20-somite stage, nz171 embryos started to exhibit developmental delay, which became increasingly apparent until homozygotes arrested in development at around 35 h.p.f.(Fig. 1B,C). nz171 embryos develop no further, and die at around 3 days postfertilization.

Giemsa-stained sections of nz171 mutants at the 20-somite stage and older revealed cells that contained abnormal condensed chromosomes, consistent with abnormal mitoses (Fig. 2G,H). To test whether developmental arrest in nz171 mutants is caused by a defect in cell division, we monitored cells in S phase with BrdU incorporation, and cells in M phase with an antibody to phosphorylated histone H3 (pH3). We found no significant difference in the number of cells in S phase between wild-type siblings and nz171 mutants (Fig. 2A,B); however, there were many more cells in mitosis. Cells in M phase started to accumulate from the 14-somite stage, well before a developmental delay is evident (Fig. 2C,D). By the 20-somite stage, there was a marked overrepresentation of cells in M phase in nz171 mutant embryos(Fig. 2E,F). These results are consistent with a block at M phase, in which nz171 mutant embryos take longer to complete mitosis and eventually arrest with all cells blocked in M phase. Giemsa staining of sections revealed that chromosomes in developmentally arrested embryos are condensed and disorganized(Fig. 2G,H). High power microscopy of pH3⁺ cells indicated that the condensed, disorganized chromosomes are mitotic (Fig. 2H-J). Cells arrested in mitosis probably undergo apoptosis, as revealed by TUNEL. Cell death was particularly prevalent in the ICM and nervous system (Fig. 2K,L).

We mapped the nz171 mutation to chromosome 16 between simple sequence length polymorphism (SSLP) markers z25049 and z51029(Fig. 3A). The cohesin subunit rad21 also maps to this region, and because of the observed mitotic defect, it was a strong candidate gene for the mutation. rad21 cDNA was isolated from nz171 homozygotes, sequenced, and a mutation identified in the coding region of the gene (nucleotide 829, exon 8) changing codon 277 from GGA, specifying glycine, to the stop codon TGA (G277X)(Fig. 3B and see Fig. S2 in the supplementary material). An antibody directed against the C-terminal region of human RAD21 detected a specific band in wild-type, but not in mutant embryos(Fig. 3C). An insertional mutant previously mapped to the rad21 locus(Amsterdam et al., 2004) has a severe early embryonic phenotype (ZFIN ID: ZDB-LOCUS-041006-4) similar to that of nz171, consistent with these defects affecting the same gene.

In wild-type embryos, rad21 mRNA was detected by RT-PCR at the oocyte stage (Fig. 3D),indicating that the transcript is maternally deposited. Whole-mount in situ hybridization analysis showed that rad21 was expressed throughout the embryo in early embryogenesis (Fig. 3E,F; data also available on ZFIN, http://zfin.org/). Expression in the brain and posterior tail regions at 26 h.p.f. was particularly robust, most likely because these are areas of active cell division. By contrast, rad21 transcription was dramatically downregulated in nz171 mutants (Fig. 3E, Fig. 8D),probably owing to nonsense-mediated mRNA decay(Chang et al., 2007). At 48 h.p.f., rad21 was strongly expressed in discrete areas of the brain,the mandibular cartilage and branchial arches, the otic vesicle and developing pectoral fins (Fig. 3F). Although some cells in these regions would be proliferating rapidly (e.g. the pectoral fins) it is surprising that this expression pattern is so specific. This might reflect a tissue-specific function for Rad21 that is not related to the cell cycle.

To provide further evidence that we had found a mutation in zebrafish rad21, we designed antisense morpholino oligonucleotides (MOs)directed against the rad21 transcript (see Fig. S2 in the supplementary material). MO-injected embryos (termed `morphants')phenotypically resembled nz171 mutants, had excess pH3⁺ cells and showed dramatic reduction in hematopoietic runx1 expression(Fig. 4A-I,M). Furthermore, we were able to rescue nz171 mutants with rad21 mRNA: nz171 mutants approached wild-type morphology at 48 h.p.f. following injection of zebrafish rad21 mRNA (Fig. 4J,K). In addition, hematopoietic runx1 expression was rescued (Fig. 4N compared with L), as were mitoses (see Table S1 in the supplementary material;data not shown). By contrast, rad21 mRNA containing the G277X mutation (zrad21^G277X) was unable to rescue nz171 mutants(see Table S1 in the supplementary material). Significantly, runx1expression was also rescued by human RAD21 mRNA(Fig. 4O and see Table S1 in the supplementary material), indicating conservation of Rad21 function in both the cell cycle and gene expression through evolution. These results provide conclusive evidence that the nz171 mutation affects the rad21 locus. We have named our mutant allele rad21^nz171accordingly.

Initial experiments revealed that runx1 expression in rad21^nz171 mutants was absent from the ALM and PLM. To determine the exact nature of the runx1 expression defect, we performed a time-course analysis of runx1 expression in rad21 morphant and rad21^nz171 mutant embryos(Fig. 5). We found that runx1 expression was never initiated in the PLM of morphants and mutants (Fig. 5A), but was still expressed in most RB cells (Fig. 5A,B, Fig. 6A). runx1 mRNA was also deficient in the ALM of rad21^nz171 mutants, although the occasional runx1-positive cell was found(Fig. 5C). Loss of rad21 did not prevent later initiation of neuronal runx1expression, as evidenced by expression of runx1 in neuronal regions of 28 h.p.f. rad21^nz171 mutants (although at reduced levels compared with wild type; Fig. 5D). Since the dorsal aorta does not form in rad21^nz171 mutants(Fig. 2H), later expression of runx1 there could not be assessed.

In early rad21^nz171 embryos, we observed normal expression of many other early transcription factors essential for hematopoiesis, such as cbfb (Fig. 1I) scl (also known as tal1 - ZFIN) and pu.1 (also known as spi1 - ZFIN)(Fig. 6A,B), cmyb, cebpa,drl and gata2 (see Fig. S1 in the supplementary material). The only other early hematopoietic transcription factor affected in rad21^nz171 was gata1, which is reduced to about half its normal expression in 10-somite embryos(Fig. 6A). A time-course analysis of gata1 expression revealed that its onset is slightly delayed in rad21-compromised embryos, with levels consistently reduced during early embryogenesis (see Fig. S3 in the supplementary material). This might, at least in part, be due to the positive autoregulation of gata1 (Kobayashi et al.,2001) by a transcriptional regulatory complex that also contains Runx1 (Elagib et al., 2003; Waltzer et al., 2003).

Normal expression of early blood markers in rad21^nz171embryos occurred at the same stage that runx1 expression is lost. Therefore, loss of runx1 is not due to a developmental delay, or loss of hematopoietic precursors in the regions where runx1 is normally expressed.

In contrast to markers of early hematopoiesis, we determined that later expressed markers, such as cebpg, lyz, hbbe3(Fig. 6C,D), lcp1 and mpx (data not shown) were severely reduced or entirely absent in 24 h.p.f. rad21^nz171 mutants. A summary of the blood defects present in rad21^nz171 is shown in Fig. S1 (see Fig. S1 in the supplementary material). Injection of 200 pg runx1 mRNA into rad21^nz171 homozygotes at the 1-cell stage rescued expression of lyz (Fig. 6D, n=21/23), demonstrating an ability for Runx1 to exert function in early myeloid cells (see Discussion). It was not possible to analyze the impact of early runx1 loss on definitive hematopoiesis in rad21^nz171 because of the developmental arrest by 35 h.p.f.

By 48 h.p.f., rad21^nz171 embryos have arrested in development with extensive mitotic defects(Fig. 2G-J). We used immunofluorescence and confocal microscopy to examine cell cycle defects in rad21^nz171 embryos during development. At the 14-somite stage, rad21^nz171 embryos can be identified by greater numbers of pH3⁺ mitotic cells(Fig. 2D). Interestingly, Rad21 protein was still detectable by immunofluorescence in 14-somite rad21^nz171 embryos(Fig. 7A), a stage when runx1 and runx3 expression was abrogated. Rad21 was associated with the nuclei of non-mitotic cells, and became displaced from nuclei in M phase (Fig. 7A). Although there were greater numbers of M phase cells in rad21^nz171 embryos(Fig. 2C,D), mitotic cells appeared normal. The increase of M phase cells in rad21^nz171 mutants at this stage probably reflects an increase in the length of time taken for cells to complete mitosis, as the available pool of maternal Rad21 becomes progressively depleted. From 29 h.p.f., Rad21 protein is no longer detected in rad21^nz171embryos (Fig. 3C). In 48 h.p.f. rad21^nz171 embryos, multiple cells in a field had arrested in mitosis with highly abnormal spindles(Fig. 7B,C). Condensed chromosomes were spread throughout these cells with clumping at the poles (see also Fig. 2H,J). These observations were reiterated in rad21 morphants(Fig. 7D). Our data are consistent with previous findings that Rad21 function in the cohesin complex is essential for holding sister chromatids together prior to anaphase(Nasmyth and Haering,2005).

We next asked if Rad21 is operating as part of the cohesin complex in the regulation of Runx gene expression. We used MOs to knock down the function of Smc3, another integral subunit of the cohesin complex (see Fig. S4 in the supplementary material). smc3 morphants appeared morphologically similar to rad21 morphants and the rad21^nz171mutant (see Fig. S4 in the supplementary material). We observed mitotic cells in smc3 morphants that had condensed chromosomes spread throughout the cell, similar to cells lacking Rad21(Fig. 7E), indicating a similar role in chromosome cohesion. Lagging chromosomes and ectopic location of chromosomes at the poles were frequently observed. Furthermore, smc3morphants lacked early runx1 and runx3 expression(Fig. 7F-I). By contrast, a hydroxyurea and aphidicolin S-phase block of the cell cycle had no effect on runx1 expression (data not shown). Since cells in rad21^nz171 homozygotes successfully complete mitosis at the 14-somite stage, it is unlikely that loss of Runx gene expression at this time is due to a cell cycle block or the activation of cell cycle checkpoints. We therefore believe that the loss of Runx gene expression in rad21^nz171 homozygotes is directly related to a reduction in cohesin. Taken together, our data indicate that the cohesin complex is necessary for normal expression of Runx genes, in addition to its function in sister chromatid cohesion.

Since cells lacking Rad21 cannot divide properly(Sonoda et al., 2001), a maternal complement of rad21 mRNA or Rad21 protein is likely to sustain proliferation until cells in rad21^nz171 embryos eventually arrest in mitosis on depletion of Rad21 protein. In support of this notion, we found that rad21 mRNA is maternally inherited(Fig. 3D). Furthermore, Rad21 protein was detected by immunoblotting in 10-somite and 14-somite rad21ATGMO morphants, and in 18-somite rad21^nz171homozygotes (Fig. 8A,B).

If cells in early rad21^nz171 mutants contain Rad21 protein and continue to proliferate (Fig. 7A), but are unable to express runx1 or runx3(Fig. 1G,H, Fig. 5A-C), then the correct regulation of Runx genes might depend on the dose of Rad21 protein. To determine whether Rad21 protein levels change as a result of gene dose, we quantified Rad21 protein in a rad21^nz171 sibling pool (of which two out of three are heterozygotes) in comparison with wild-type embryos. Rad21 protein was reduced by 40%(Fig. 8C) commensurate with a reduction in rad21 mRNA (Fig. 8D, top left) in rad21^nz171 siblings. Using quantitative RT-PCR, we compared runx1 expression in rad21^nz171 mutants (-/-) and siblings (+/- and +/+) with wild-type (+/+) embryos at 24 h.p.f. We found that runx1transcription is moderately reduced in rad21^nz171 siblings relative to wild-type, with a further reduction in mutants(Fig. 8D, top right). These results indicate that early runx1 expression is sensitive to changes in the dose of rad21. Interestingly, we detected no difference in runx1 levels between +/+ embryos and rad21^nz171siblings at 48 h.p.f. (data not shown), indicating that later expression of runx1 is probably not Rad21-dependent.

rad21^nz171 homozygotes have defects in neuronal development during early embryogenesis (J.A.H. and J.K-H.H., unpublished) and rad21 is specifically expressed in restricted regions of the brain at 48 h.p.f. (Fig. 3F). To determine if changes in the dose of rad21 can affect the expression of neuronal genes in older embryos, we used quantitative RT-PCR to monitor expression of selected neuronal genes in 48 h.p.f. embryos. We found that expression of proneural genes ascl1a and ascl1b, which are strongly expressed in the brain during embryogenesis(Allende and Weinberg, 1994),was severely reduced in rad21^nz171 mutants and significantly reduced in siblings (Fig. 8D, lower graphs). Our data indicate that although rad21^nz171 siblings appear to grow and develop normally,they are unable to express wild-type levels of downstream genes. This finding points to a regulatory function for cohesin that can be separated from its cell cycle role.

To understand how Rad21/cohesin might influence Runx gene expression, we investigated whether loss of rad21 affects pathways known to be upstream of runx1. The cdx (hox) pathway specifies blood development from mesoderm, and is essential for runx1expression in embryogenesis (Davidson et al., 2003; Davidson and Zon,2006). We determined that the expression of cdx4 and its downstream hox targets (hoxa9a, hoxb4, hoxb6b, hoxb7a) were unaffected in 10-somite rad21^nz171 homozygotes (data not shown). The Notch signaling pathway is upstream of definitive runx1expression in zebrafish (Burns et al.,2005). Interestingly, we found that several genes induced by Notch signaling are downregulated in rad21^nz171 embryos (J.A.H. and S.H.A., unpublished results). However, consistent with previous data(Burns et al., 2005; Gering and Patient, 2005), we found that early expression of runx1 is not Notch dependent (data not shown), thereby eliminating this pathway as an intermediate between cohesin and early runx1 expression. In summary, two previously characterized pathways shown to contribute to runx1 expression do not mediate the cohesin regulatory function. Therefore cohesin appears to have a novel role in the regulation of early runx1 expression.

The impact of cohesin loss on runx1 expression appears to be extraordinarily specific. Other zebrafish mutations affecting early runx1 transcription also perturb the expression of multiple hematopoietic transcription factors (e.g. spt, kgg)(Hsia and Zon, 2005). Therefore, rad21^nz171 appears to be the first mutant that affects early expression of runx1 alone. Although most other early hematopoietic markers are expressed normally in rad21^nz171, early myelopoiesis is deficient and hbbe3 expression is reduced. By 24 h.p.f., the cell cycle is severely affected in rad21^nz171(Fig. 2) and therefore hematopoietic progenitors might be unable to divide to form differentiated progeny. In addition, cell death in the ICM almost certainly contributes to a reduction in differentiated cells (Fig. 2L). However, live hematopoietic progenitors remain in the ICM of rad21^nz171 mutants, as indicated by robust scland gata2 expression at 24 h.p.f.; furthermore, inappropriate maintenance of scl expression in the ICM at 48 h.p.f. points to the persistence of immature precursors that cannot differentiate (see Fig. S1 in the supplementary material). Loss of differentiated cells in rad21^nz171 might be due in part to loss of runx1and reduction in gata1 mRNA. Rescue of lyz⁺ cells by runx1 mRNA indicates that either, (1) runx1 is necessary for early myelopoiesis and its restoration rescues differentiated cells, or(2) expression of runx1 is able to overcome a developmental block in the myeloid precursors of rad21^nz171 by driving more cells toward a myeloid fate at an earlier time. In support of the latter, we observed an increase in the number of lyz⁺ cells in siblings of runx1-injected rad21^nz171 crosses(n=41/41, data not shown). Unfortunately, the cell cycle block in rad21^nz171 prevented an analysis of effects on definitive hematopoiesis.

Our results add to increasing evidence that chromatin-modifying proteins can have specific roles in hematopoiesis. In previous studies, a nucleosome assembly protein NAP1L was shown to operate upstream of scl in Xenopus hematopoiesis (Abu-Daya et al., 2005). Furthermore, Brg1 (a SWI/SNF subunit) appears to have a distinct role in the activation of the β-globin locus in erythropoiesis(Bultman et al., 2005). The involvement of trithorax (Ernst et al.,2004a; Ernst et al.,2004b) and polycomb (Lessard and Sauvageau, 2003; Lessard et al., 1999) group members as epigenetic regulators of hematopoietic stem cell development is relatively well characterized.

We report the first direct example of cohesin-dependent gene regulation in a vertebrate. Previous studies in Drosophila implicated cohesin loading protein, Nipped-B, and cohesin subunits in regulation of the cut and Ultrabithorax loci(Dorsett et al., 2005; Rollins et al., 1999). In C. elegans, the cohesin-loading factor MAU-2 (the Scc4 ortholog) is essential for axon migration during development(Seitan et al., 2006). Our observations of neuronal abnormalities and altered neuronal gene expression in rad21^nz171 embryos, together with the C. elegansdata, are consistent with the idea that chromosome cohesion proteins have specific roles in neuronal development.

Significantly, we uncovered a potent dose effect of Rad21 levels on gene expression (Fig. 8). Our results raise the possibility that there is a threshold level of Rad21, below which cell proliferation can be sustained, but gene expression is compromised. Perhaps the mitotic function of cohesin is prioritized at the expense of its alternative (potentially regulatory) functions in a depleted cohesin pool.

It is tempting to speculate that loss of early runx1 and runx3 expression in rad21^nz171 is causally related to the connection between the Runx family and the cell cycle. Runx protein levels are dynamically regulated during the cell cycle; e.g. expression of Runx2 oscillates during the cell cycle of MC3T3 osteoblasts (Galindo et al.,2005). In cell lines, Runx1 levels were also shown to oscillate in a cell cycle-dependent manner: Runx1 protein levels increase during S phase and G2 (Bernardin-Fried et al.,2004) and at the G2-M phase transition, Runx1 is degraded by the anaphase-promoting complex (Biggs et al.,2006; Wang et al.,2006). Cohesin becomes stably associated with chromatin during S and G2 (Gerlich et al., 2006),and perhaps this is necessary for runx1 expression in a particular developmental context. Runx1 degradation is also concomitant with cohesin cleavage at G2-M.

Runx proteins appear to have cell cycle-specific functions. Overexpression of Runx1 causes a shortening of G1 phase(Strom et al., 2000) and Runx1 physically interacts with cyclin D3 to repress its own transcription(Peterson et al., 2005). During mitosis, Runx2 selectively regulates specific target genes(Young et al., 2007b), and represses transcription of ribosomal RNA genes(Young et al., 2007a). These authors also reported similar unpublished observations for Runx1. Therefore,Runx proteins might regulate cell growth through control of ribosome biogenesis, and it was proposed that by this mechanism Runx proteins might coordinate cell proliferation and differentiation. Clearly, Runx proteins provide a mechanistic link between the cell cycle and development. Therefore a mechanism by which Runx gene expression is coordinated with the cell cycle would make sense. Cohesin is an integral part of the cell cycle machinery, and is therefore a good candidate to participate in cell cycle-dependent gene regulation.

There are a number of human developmental disorders associated with loss of sister chromatid cohesion, including CdLS(Krantz et al., 2004; Musio et al., 2006; Strachan, 2005). This raises the interesting possibility that such developmental disorders might be contributed to by a reduction in Runx gene expression in embryogenesis, as a result of reduction in cohesin function. CdLS patients present with neurodevelopmental, gastrointestinal and skeletal abnormalities(Strachan, 2005). The development of each of these systems depends on the proper regulation of Runx proteins, which are themselves dose-sensitive in function(Blyth et al., 2005). In summary, our findings provide strong evidence for a novel developmental function for cohesin. The next challenge will be to determine exactly how cohesin contributes to regulation of gene expression and developmental pathways.

The authors thank Wendy Flühler for excellent technical assistance,Anassuya Ramachandran for help with quantitative PCR, Alhad Mahagaonkar and Peter Cattin for zebrafish husbandry, and Latifa Khan for help with screening the F1. We also thank Debbie Yelon, Didier Stainier, Joan Heath and members of the Stainier, Heath and Lieschke laboratories for screening and mapping advice, and Helena Richardson for critical reading of the manuscript. This work was supported by Health Research Council of New Zealand grant 04/120 to P.S.C., J.A.H., and K.E.C.