Sox9 is required for ES cell differentiation into cartilage nodules in vitro Our results demonstrate that in vitro differentiation of Sox9^-/- ES cells into chondrocytes is disrupted at a stage characterised by the appearance of typical round-shaped chondrocytes orga- nized in distinct nodules and expressing a high level of type II collagen. Sox9-/- cells fail to form these cartilage nodules. In line with this, it has been demonstrated that Sox9-deficient cells could not differentiate into mature chondrocytes in mouse Sox9-/- chimeras and teratomas (Bi et al., 1999). Furthermore, condi- tional null mutant mice lacking Sox9 in mesenchymal cells of limb buds were not able to form cartilage and bone in limbs (Akiyama et al., 2002). Since we found in previous studies that the formation of cartilage nodules in differentiating EBs can be induced after application of growth factors of the transforming growth factor Я- family (Kramer et al., 2000) we applied BMP-2, TGF-β₁ and TGF-β₃ to differentiating Sox9^-/- EBs. However, we did not detect any cartilage nodule in these EB outgrowths (data not shown) indicat- ing that loss of Sox9 function cannot be rescued by these chondrogenic factors in vitro. In Sox9+/- EB outgrowths, cartilage nodules were present, but the number of nodules appeared to be reduced compared to the wildtype, indicating that Sox9 gene dosage is important for proper cartilage differentiation. This agrees with the observation that cartilage structures are defective and hypoplastic but not completely absent in CD patients and Sox9+/- mice (Houston et al., 1983; Bi et al., 2001). Taken together, this demonstrates that Sox9 plays an essential role during chondrogenesis both in vivo and in vitro. It has been proposed that Sox9 plays an inhibitory role for the switch from prehypertrophic to hypertrophic chondrocytes, be- cause Sox9 is completely switched off in hypertrophic chondro- cytes in vivo (Zhao et al., 1997; Bi et al., 2001). Furthermore, in prehypertrophic chondrocytes of mice lacking the receptor for the parathyroid hormone-related peptide, Sox9 phosphorylation is abolished and these mice show accelerated differentiation of hypertrophic chondrocytes (Lanske et al., 1996; Huang et al., 2001). Our data suggest that Col10a1 expression and the number of Col10a1- expressing cells increase in Sox9^+/- nodules. These results indicate that a reduced level of Sox9 may promote the formation of hypertrophic chondrocytes. In line with this, Sox9^+/- mice showed an enlarged zone of chondrocyte hypertrophy in the growth plates of the long bones (Bi et al., 2001).

Formation of pre-cartilage condensations is not affected in Sox9-/- EBs We found that during ES cell differentiation in vitro, Sox9- deficient cells form early pre-cartilage condensations which ex- press Sox5, Sox6, scleraxis and N-cadherin, markers character- istic for such pre-cartilage condensations (Oberlender and Tuan, 1994; Lefebvre et al., 1998; Brown et al., 1999). Furthermore, the condensations strongly bind the lectin PNA (DeLise et al., 2000). These results were not only obtained with both subclones from the Sox9^-/- clone D4D12, lacking exon 2, but also with the indepen- dent Sox9^-/- clone 2A5-40, lacking exons 2 plus 3. In mouse chimeras, Sox9^-/- cells were located adjacent to condensing wildtype mesenchymal cells in 11.5 and 12.5 d p.c. embryos and did not take part in the formation of mesenchymal condensations (Bi et al., 1999). Moreover, conditional null mutant mice, in which Sox9 was inactivated in mesenchymal cells of limb buds before these cells condense, were no longer able to form mesenchymal condensations (Akiyama et al., 2002). These results suggested that condensation formation is a cell-autono- mous process under the control of Sox9. On the other hand, zebrafish with homozygous null mutations in the sox9a gene, an ortholog of the mammalian Sox9 gene, are able to form mesen- chymal condensations in the first two pharyngeal arches (Yan et al. , 2002). We have shown in the present study, that a homoge- neous population of Sox9-deficient cells is able to form pre- cartilage condensations and express early molecular markers in vitro . Thus, our data indicate that the function of Sox9 during condensation formation is rather not cell-autonomous. In vivo, development is not only temporally controlled as in EBs, but is also spatially controlled by a combination of distinct signaling molecules present at defined concentrations. Because morphogenetic development is not possible within EBs, such spatially controlled signals might be lacking, resulting in conden- sation of mesenchymal cells in Sox9^-/- EBs. In line with this, a variability of morphogenetic signals around pre-cartilage con- densations in EB outgrowths might influence their cellular fate. In fact, we found that in wildtype EBs only some of the condensa- tions develop into cartilage nodules, and the mean number of scleraxis-positive condensations does not decrease significantly during culture. Thus, the in vitro system offers the possibilty to analyze the potency of cell-autonomous differentiation effects. The formation of specifically shaped mesenchymal condensa- tions in vivo may be regulated by antagonistic factors produced by ectodermal or non-condensing mesenchymal cells located close to the place of the condensations (Zanetti and Solursh, 1986). For example, in limb buds, the size and shape of mesenchymal condensations is controlled by inhibitory factors produced by ectodermal cells such as FGF2 and FGF8 (Moftah et al., 2002). Sox9 could be an antagonist of such condensation-inhibiting factors, which are expressed in vivo by the overlaying ectoderm. The absence of Sox9 would then result in the complete loss of condensations in vivo. In contrast, in the EB in vitro differentiation system, the release of such condensation-inhibiting factors from adjacent tissue can not occur and therefore condensations can form even in the absence of Sox9. Similarly, it has been found that scleraxis null mutant embryos fail to form mesoderm, whereas in EBs of scleraxis^-/- ES cells, mesodermal markers were ex- pressed at a similar level as in wildtype EBs (Brown et al., 1999) indicating that the role of scleraxis during mesoderm formation is not cell-autonomous but depends on the environment. Another example are mice lacking a transcription factor, the serum re- sponse factor (Srf). These Srf^-/- mice stop developing at the onset of gastrulation and do not form mesoderm (Arsenian et al., 1998). However, Srf^-/- ES cells differentiated in vitro into mesodermal cell types although this process was impaired in vivo (Weinhold et al. , 2000) suggesting that the function of Srf to promote meso- derm formation is non-cell-autonomous. Our data indicate that a non cell-autonomous function may also apply to Sox9 regarding the formation of pre-cartilage condensations.

Loss of Sox9 affects the level of but does not completely abolish Col2a1 expression in vitro Expression studies in vivo suggested that Col2a1 is a target for Sox9, because Sox9 and Col2a1 are coexpressed in cartilage primordia throughout the developing skeleton and in other devel- oping cartilage structures during embryogenesis (Zhao et al., 1997; Ng et al., 1997). It has also been shown that Sox9 binds to specific sequence elements of the Col2a1 enhancer and directs chondrocyte-specific Col2a1 expression, both in transient trans- fection experiments and in transgenic mice (Lefebvre et al., 1996; Bell et al., 1997; Lefebvre et al., 1997; Zhou et al., 1998). In mouse chimeras, Sox9^-/- cells did not express Col2a1, and in Sox9-/- teratomas, type II collagen was not detectable in any cell type (Bi et al., 1999). In contrast, coexpression studies of Sox9 and Col2a1 in developing wild type mouse embryos revealed that Col2a1 was expressed in several nonskeletal tissues which are negative for Sox9 (Ng et al., 1997), indicating that differentiating cells are able to express Col2a1 in the absence of Sox9. Furthermore, type II collagen expression in cultured human articular chondrocytes does not correlate with the level of Sox9 expression (Aigner et al., 2003). We found that Sox9-deficiency did not result in complete abolishment but obvious downregulation of Col2a1 expression in differentiating EBs, as shown by real-time RT-PCR and by detection of type II collagen fibres in Sox9^-/- EB out- growths. One possible explanation for this unexpected result could be that we generated a partially functional Sox9 protein by our gene targeting strategy. This can be ruled out, as we did not detect any Sox9 protein in cultured Sox9^-/- EBs. Furthermore, it has been shown recently that no functional Sox9 protein could be detected after conditional inactivation of Sox9 in the lung using the same targeting strategy resulting in Sox9-del^ex2,3 alleles (Perl et al. , 2005). Moreover, even if a truncated protein was still produced in the knock-out lines this mutant Sox9 protein would lack a functional DNA binding domain and the C-terminal transactivation domain (Sьdbeck et al., 1996) and would thus not be able to function as a transcription factor. Another explanation would be compensation of Sox9 function in vitro by a protein with an overlapping function. The transcription factors Sox5 and Sox6 would be candidates. Both were ex- pressed in Sox9-deficient condensations in vitro and might up- regulate Col2a1 expression at least to moderate expression levels. Such a compensatory mechanism may depend on the prior formation of mesenchymal condensations. This agrees with the observation that in conditional null mutant mice, expression of Col2a1 as well as Sox5 and Sox6 was inhibited when Sox9 was inactivated in mesenchymal cells before condensations had been formed (Akiyama et al., 2002). In conclusion, this in vitro study unravels a mechanistic insight into the function of Sox9 during chondrogenic differentiation. We found that in contrast to the terminal step of differentiation characterized by the formation of cartilage nodules, the early step of pre-chondrogenic differentiation, the formation of pre-cartilage condensations, remained almost unaffected after loss of Sox9 function in vitro. In contrast, a block of this early differentiation step has previously been demonstrated in vivo. This indicates that the function of Sox9 during this process is not cell-autonomous. Thus, the in vitro differentiation of embryonic stem cells is a useful approach to bring new important insights into complex develop- mental processes.