Developing vertebrate limbs initiate proximo-distal patterning by interpreting opposing gradients of diffusible signaling molecules. We report two thresholds of proximo-distal signals in the limb bud: a higher threshold that establishes the upper-arm to forearm transition; and a lower one that positions a later transition from forearm to hand. For this last transition to happen, however, the signal environment seems to be insufficient, and we show that a timing mechanism dependent on histone acetylation status is also necessary. Therefore, as a consequence of the time dependence, the lower signaling threshold remains cryptic until the timing mechanism reveals it. We propose that this timing mechanism prevents the distal transition from happening too early, so that the prospective forearm has enough time to expand and form a properly sized segment. Importantly, the gene expression changes provoked by the first transition further regulate proximo-distal signal distribution, thereby coordinating the positioning of the two thresholds, which ensures robustness. This model is compatible with the most recent genetic analyses and underscores the importance of growth during the time-dependent patterning phase, providing a new mechanistic framework for understanding congenital limb defects.
Development of the main axis of vertebrate limbs is a paradigm for segmental specification in a growing structure. The limb bud arises from the lateral plate as a bulge of mesenchymal cells encased within an ectodermal hull, which progressively grows from the embryo flank. Vertebrate limbs develop their three main proximo-distal (PD) segments (upper arm/leg or stylopod, lower arm/leg or zeugopod, and hand/foot or autopod) in a proximal-to-distal sequence. An undifferentiated region at the distal end of the limb bud contributes cells that differentiate into progressively more distal segments until all structures are generated (reviewed by Tabin and Wolpert, 2007). The ‘differentiation front’ model (Tabin and Wolpert, 2007) proposes that a given segment is correctly produced only when enough cells pre-specified to form that segment leave the undifferentiated distal region and incorporate into the limb axis (Galloway et al., 2009).
At the late limb bud stage, each of the PD segments expresses specific homeobox genes; the stylopod expresses Meis1 and Meis2, the zeugopod Hoxa11 and the autopod Hoxa13 (Mercader et al., 2009; Nelson et al., 1996; Yokouchi et al., 1991). However, the expression of these so-called PD markers is dynamic and sequential during early stages. Initially, Meis1/2 are expressed throughout the limb bud, and then are downregulated in the distal region, where subsequent activation of Hoxa11 takes place. Later, Hoxa13 expression is activated in a small posterodistal domain and then expands along the AP and the PD axes, overlapping for some time with Hoxa11 expression; finally, Hoxa11 is downregulated distally. The timing and extent of the transcriptional activation and repression of this set of genes allows inference of the degree of limb distalization (Tabin and Wolpert, 2007), and thus mRNA in situ hybridization is the technique commonly used to interrogate their expression in different experimental settings.
Diffusible molecules play a key role during limb development. The pool of distal undifferentiated cells responsible for limb generation is maintained by fibroblast growth factor (FGF) and Wnt signals produced by a distal epithelial structure called the apical ectodermal ridge (AER) (ten Berge et al., 2008). For their part, the somites and the pre-bud lateral plate mesoderm (LPM) express high levels of RALDH2, one of the enzymes necessary for synthesizing retinoic acid (RA) from vitamin A, leading to high RA signaling within the whole early limb bud (Dollé et al., 2010; Mic et al., 2004). Upon limb induction, RA synthesis ceases in the limb bud and RA degradation starts in its distal region upon activation of the RA-degrading enzyme CYP26B1 (Yashiro et al., 2004). The combination of limb bud growth and the inverted distribution of RA synthesis and degradation leads to a PD gradient of RA signaling (Mic et al., 2004; Yashiro et al., 2004). Although it is widely accepted that AER signals, especially FGFs, have distalizing effects, mainly by inhibiting the proximal influence from the flank (Mariani et al., 2008; Mercader et al., 2000; Roselló-Diez et al., 2011; Roselló-Diez and Torres, 2011), the relevance of the RA signaling gradient is debated. Several studies in chicken and mouse embryos have shown that RA from the flank has limb-proximalizing effects that are counteracted by AER-FGFs (Cooper et al., 2011; Mariani et al., 2008; Mercader et al., 2000; Roselló-Diez et al., 2011). These observations support the two-signal model (Mercader et al., 2000), according to which PD specification depends on the opposed action of proximal RA and distal FGF signals. However, the apparently normal Meis gene expression and PD hindlimb patterning in mice deficient for the enzyme RDH10 (part of the RA synthesis pathway) have challenged the view of RA as an important signal for limb patterning (Cunningham et al., 2013). Further studies, however, indicate that the mutant mouse used in that study (T-rex) bears a hypomorphic allele that allows embryo development up to embryonic day (E) 13-E14 (Sandell et al., 2012; Sandell et al., 2007), whereas Raldh2-null embryos (which are expected to have a greater deficiency of RA) only reach E9 (Niederreither et al., 1999). Therefore the T-rex mutants in fact contain functional RA levels, and their validity for studying the role of RA in limb development is thus questionable (see also Discussion).
In conclusion, although it is obvious that some further studies are required, the two-signal model is the most compatible with the accumulated experimental evidence, and RA remains the best candidate for the proximal patterning signal. However, although this mechanism clearly controls the early stylopod-zeugopod transition, its involvement in further PD transitions is less explored and was the main objective of this study.
There are two PD signaling thresholds
The simplest signal-based model for PD patterning would be the specification of stylopod by high levels of RA versus FGF, the zeugopod by intermediate levels, and the autopod by low or null levels. To test this model, we first studied the effect of CYP26B1 inhibition on Hoxa13 expression. Treatment of early chick wing buds with beads soaked in the CYP26 inhibitor R116010 (Armstrong et al., 2007) delayed both the onset of Hoxa13 expression (Fig. 1A, n=8/8 with severely-reduced or no expression) and the expansion of its initial expression domain (Fig. 1B, n=8/10), a result similar to that observed in Cyp26b1 mutants (Yashiro et al., 2004). The fact that this process is only delayed, and not blocked, could be due either to bead exhaustion or to the endogenous degradation of RA taking place eventually even in absence of CYP26B1 activity. In any case, this short-term treatment does not affect Meis1 or Hoxa11 expression (Fig. 1C,D, n=0/4 each), arguing against a non-specific effect of the chemical inhibitor, and suggesting that there are two distinct thresholds of RA signaling: while the Meis-Hoxa11 transition depends mainly on the inhibition of RA signaling by FGF (Cooper et al., 2011; Mariani et al., 2008; Mercader et al., 2000; Roselló-Diez et al., 2011), the activation of Hoxa13 at the appropriate time requires further reduction of RA signaling by active RA degradation.
MEIS controls RA degradation via CYP26B1
We noticed that the phenotype observed upon CYP26 inhibition (Fig. 1) resembles that of Meis1 misexpression in the distal limb bud, namely a distal shift in Hoxa13 activation (Mercader et al., 1999; Mercader et al., 2009). As the mechanisms by which MEIS proteins affect the distal PD transition are unknown, we set out to investigate this shared phenotype.
To gain insight into the role of MEIS, we turned to a conditional misexpression mouse model. We generated a mouse knock-in line (R26loxP-STOP-loxP-Meis2a-IRES-eYFP; R26RM2, hereafter) in which Meis2a and eYFP expression is activated by Cre activity (supplementary material Fig. S1). We then injected 4-hydroxy-tamoxifen (4HT) at E8.3 into crosses with the HoxB6-CreER line, in which tamoxifen-inducible Cre recombinase expression is evident in the LPM and limb bud precursors as early as E8.5 (Nguyen et al., 2009). Activation of the transgene took place in the whole hindlimb bud (HL) and the posterior half of the forelimb bud (FL) at E10.5 (Fig. 2A,A′, n>100). In situ hybridization revealed delayed onset and impaired expansion of the Hoxa13 expression domain in Meis2a-misexpressing limbs (Fig. 2B-E′, n>30 experimental, 15 control embryos). Meis2a misexpression did not ectopically activate Meis1 (Fig. 2H′′), and did not affect the timing of Hoxa11 activation (not shown). However, Hoxa11 downregulation in the distal limb bud was delayed (Fig. 2D′,E′, n=8 experimental, 4 control embryos), so that autopod precursor cells misexpressing Meis2a had a HoxA expression profile equivalent to that of the wild-type zeugopod precursors at that stage. Given that Hox11 paralog function is unimportant for Hoxa13 activation (Davis et al., 1995), but that Hox13 proteins are essential for repression of Hoxa11 in the prospective autopod (Sheth et al., 2013), the simplest explanation for these results is that MEIS represses Hoxa13 and that this results in failed inactivation of Hoxa11. A similar though less penetrant effect was obtained when Meis2a misexpression was driven by the ShhGfpCre line (Harfe et al., 2004), the relative weakness of the effect possibly being due to the late activation of the transgene compared with the HoxB6-CreER line (supplementary material Fig. S1). Importantly, in situ hybridization on adjacent tissue sections showed that Hoxa11 or Hoxa13 expression is affected only in cells expressing the transgene, indicating that the effect of Meis2a misexpression is cell-autonomous (Fig. 2H,H′,H′ for HoxB6-CreER and supplementary material Fig. S1 for ShhGfpCre).
As CYP26 inhibition (Fig. 1) and MEIS2 misexpression have similar effects on Hoxa13 expression, we compared Cyp26b1 expression in control and transgenic embryos, finding that Cyp26b1 was notably downregulated by Meis2a misexpression (Fig. 2F,G, n=6). This suggests that MEIS, via inhibition of Cyp26b1 expression, maintains local RA levels high enough to impede the Hoxa11-Hoxa13 transition, an interpretation also in agreement with the observation that Cyp26b1-deficient limb buds show impaired Hoxa13 activation (Yashiro et al., 2004). To test this idea, we administered the RA antagonist (RAA) BMS493 in utero to try to rescue the Meis2 misexpression phenotype. We validated the efficacy of the RAA treatment by analyzing the expression of the RA target Rarb, which, as expected, was notably downregulated (Fig. 2I,J, n=4/4). Most importantly, we found that the RAA treatment significantly abolished Hoxa13 downregulation in a large fraction of Meis2-missexpressing cells (Fig. 2K-N, n=3/4). Collectively, these results indicate that a major cause of the impaired activation of Hoxa13 upon Meis2a misexpression is the maintenance of significant RA signaling in the distal limb bud.
Time is also necessary for Hoxa13 activation
The results so far suggested that if a zone devoid of RA signaling were prematurely created in the limb bud, Hoxa13 would be expressed precociously. To test this hypothesis, we went back to the easily accessible chick model. We treated HH19 chick wing buds, right at the Hoxa11 activation stage (Nelson et al., 1996), with RAA. Analysis at 8-10 hours post-insertion (hpi), around the onset of endogenous Hoxa13 expression, unexpectedly revealed no premature expression (Fig. 3A, n=0/6). To discard the possibility that additional signals were needed, we also applied beads soaked in the distal signals FGF8 and sonic hedgehog (SHH) (Mariani et al., 2008; Mercader et al., 2000; Probst et al., 2011), but this was again insufficient to prematurely activate Hoxa13 expression at 10 hpi (Fig. 3C, n=0/8). However, at 24 hpi, once there is a solidly established endogenous Hoxa13 domain, RAA treatment was able to expand the Hoxa13 expression domain (Fig. 3B, n=5/7). This result indicates that the establishment of the Hoxa13 expression domain can be modulated simply by reducing RA signaling, but only after endogenous Hoxa13 expression has begun. Moreover, the combined treatment with RAA, FGF8 and SHH was able to induce an area of ectopic Hoxa13 expression (Fig. 3D, n=2/4), but again only at 24 hpi, after the endogenous Hoxa13 expression domain has been established. This shows that even in conditions in which Hoxa13 can be induced ectopically, it does not appear before the endogenous expression starts.
These results support the idea that, in addition to the distal signaling environment, temporal competence is an essential factor for Hoxa13 expression. The idea that a timing mechanism is involved in limb distalization and is specifically needed for Hoxa13 activation by FGF has been proposed before (Summerbell et al., 1973; Vargesson et al., 2001), and divides into two main hypotheses: (1) cells integrate distal signals over time, activating Hoxa13 when the accumulated signal surpasses a certain threshold; (2) timing is molecularly encoded and interpreted independently of diffusible signals, in a completely cell-autonomous way.
We tested the first hypothesis by means of grafting experiments coupled to pharmacological treatments in the chick embryo. The undifferentiated distal tips (200 μm) of HH19-20 wing buds, which have just activated Hoxa11 and do not yet express Hoxa13, were grafted to the RA-rich somite region of HH20 embryos (Fig. 4A) and the expression of Meis1, Hoxa11 and Hoxa13 was examined 20 hours post-grafting (hpg). We chose to use grafts rather than directly manipulating limb buds because we found that complete inhibition of FGF signaling required the placing of several beads at a very distal location, something we could only achieve on excised limb bud tips. Although untreated transplants activated Hoxa13 expression in the distal region (Fig. 4B′′, n=2/2), grafts in which FGF signaling had been pharmacologically abolished did not (Fig. 4C′′, n=0/4). This result agrees with the rapid Hoxa13 downregulation observed upon AER removal (Vargesson et al., 2001). However, analysis of other molecular markers revealed that this situation is due to a reversion to early stages of PD patterning, in which the RA/FGF balance tilts towards RA (compare Fig. 4B-B′ with 4C-C′) (Roselló-Diez et al., 2011). Given that excess RA is sufficient for Hoxa13 repression (Fig. 1), these results could reflect the persistence of RA signaling rather than direct dependence on FGF. Therefore, to eliminate the influence of RA, we simultaneously inhibited FGF and RA signaling. In this PD-signal-free situation, molecular marker expression did not revert to that of earlier stages (Fig. 4D-D′, n=5/5) and Hoxa13 expression was detected in the transplant 20 hpg (Fig. 4D′′, n=4/5). FGF signaling is therefore required to keep RA signaling at bay and thereby allow Hoxa13 expression, but the integration of sustained FGF signaling is neither required nor instructive for Hoxa13 activation.
Histone acetylation status controls the timing of Hoxa13 activation
Temporal co-linear activation of Hox genes in the AP embryo axis correlates with progressive opening of Hox cluster chromatin (Soshnikova and Duboule, 2009), prompting us to explore a similar scenario for the limb PD axis (see also Discussion). Histone post-translational modification plays a major role in chromatin opening and therefore in allowing or blocking transcription (reviewed by Bannister and Kouzarides, 2011; Suganuma and Workman, 2011). Given that histone deacetylases (HDACs) are needed to switch chromatin to an inactive state, we tested whether continuous HDAC activity is required to keep Hoxa13 repressed in the early limb bud. We treated HH19 chick wing buds with beads soaked in the HDAC inhibitor trichostatin A (TSA) (Sekhavat et al., 2007) and analyzed Hoxa13 expression 5-6 hours later, a time window short enough to ensure that we were mainly observing direct effects and that endogenous Hoxa13 expression was not yet detectable. Vehicle-treated limb buds showed no effect on Hoxa13 expression (Fig. 5A, n=0/5), whereas most TSA-treated limb buds showed precocious expression (Fig. 5B, n=18/19). Importantly, Shh, Fgf8 and the FGF target Sprouty2 were frequently downregulated after TSA treatment (n=10/12, Fig. 5C; supplementary material Fig. S2) (Zhao et al., 2009a), which indicates that chromatin de-repression is sufficient for Hoxa13 activation in the absence of these limb bud signals, at least in the distal region of the limb (see Discussion).
To study the consequences of TSA treatment on the skeletal pattern, we let treated chick embryos develop for 6-7 days after bead insertion and compared the skeletal wing elements (humerus, ulna and central digit) with those of the contralateral wing. TSA significantly reduced the size of all segments analyzed (n=15, Fig. 5E,F) probably through its general effect on proliferation (Ocker and Schneider-Stock, 2007). We thus used the humerus (the least affected segment) as a control reference for the general effect on size. Determination of the reduction ratio (experimental versus contralateral) for each skeletal element showed that, despite the distal position of the bead, the zeugopod, but not the autopod, was significantly reduced in size with respect to the stylopod (Fig. 5F). The zeugopod reduction thus appears to result from a premature switch towards the autopod program, before zeugopod precursors have had time to fully expand and incorporate into the limb axis.
Notably, in the TSA-treated chick wing buds, Hoxa13 expression is activated only in a crescent-shaped domain at the distal-most region, despite the predicted spherical release of TSA. This again suggests that Hoxa13 activation requires, in addition to chromatin relaxation, a permissive environment, which at this stage would only be found in this distal crescent-shaped domain. Given the influence of RA levels on Hoxa13 expression (Fig. 1A,B; Fig. 3B), we investigated whether RA levels define this permissive environment. Supporting this idea, combination of RAA with TSA in the implanted beads extended the premature Hoxa13 domain to more proximal regions than seen with TSA treatment alone, and the premature activation was no longer restricted to a crescent shape but formed a sphere around the bead (Fig. 5D, n=8/16). Thus, whereas the timing of Hoxa13 activation is controlled by a chromatin accessibility mechanism, the size and shape of the Hoxa13 domain is determined by the signaling environment.
RA and limb development
Here, we report that the RA/FGF signal balance, in addition to establishing the first limb bud PD transition from stylopod to zeugopod (Cooper et al., 2011; Roselló-Diez et al., 2011), also regulates the second transition from zeugopod to autopod. However, as mentioned in the Introduction, the role of RA in PD limb patterning has been challenged by the characterization of mouse mutants deficient in the RA-synthesis enzymes Raldh2 or Rdh10 (Cunningham et al., 2013; Zhao et al., 2009b). These studies show that hindlimb (but not forelimb) patterning can proceed normally (including Meis expression) in situations in which RA signaling is not detected by a genetically encoded RA reporter (RARE-lacZ), making the case that RA is unnecessary for limb patterning or Meis expression. Raldh2 mutants, however, require administration of at least a brief pulse of RA to prevent embryos dying before the limb-bud stage (Zhao et al., 2009b), and therefore their limbs do not form without RA. Indeed, previous rescue experiments in which RA was administered at different stages and doses confirmed the special sensitivity of proximal limb segments and Meis gene expression to RA availability (Niederreither et al., 2002). Regarding the Rdh10 analysis, while Rdh10-null mutants die around E10.5-E12.5, the model analyzed by Cunningham et al. (Cunningham et al., 2013) is a hypomorph mutant called T-Rex, the lethal phase of which is E13.5-E14.5 (Rhinn et al., 2011; Sandell et al., 2012), and therefore T-Rex mutants contain significant amounts of functional RA, which were not detected by the reporter. Relevant to this discrepancy, the RA sensitivity assays aimed to calibrate the reporter (Cunningham et al., 2013) were carried out in vitro by whole embryo exposure to RA, a situation very different to in vivo RA delivery. In summary, the models that question the role of RA in limb patterning in fact contain low but functional RA levels, precluding any definitive conclusion about this matter.
The simplest explanation for the apparently conflicting results is that endogenous limb-patterning genes (e.g. Meis) have a lower in vivo RA activation threshold than the reporter. Supporting this idea, the RARE-LacZ expression border in limb buds is more proximal (i.e. closer to the RA source) than that of Meis2 (Yashiro et al., 2004). This different sensitivity would allow RA-mediated PD limb patterning in the absence of reporter activation and would also explain the contradictory results for Meis gene responses to RA in the mouse (Cunningham et al., 2013; Niederreither et al., 2002; Zhao et al., 2009b). In this scenario, the small amount of RA present in Raldh2 and Rdh10 mutants would still allow some degree of PD specification (such as activation of Meis expression), indicating a mechanism of outstanding robustness in the face of large variations in RA availability. As vitamin A is obtained from the environment, it is reasonable to suppose that such robust mechanisms have evolved.
Finally, although the proximalization of the PD molecular code in Cyp26b1 mutants is consistent with the two-signal model, the fact that chondrocyte differentiation is affected in the three limb segments, leading to very dysmorphic limbs (Yashiro et al., 2004), has sometimes been interpreted as evidence that RA is in fact a teratogen for limb development and that CYP26B1 is required to protect the limb from its action (Zhao et al., 2009b). However, the effect on chondrocyte differentiation is clearly a distinct and later effect of RA that can be uncoupled from the patterning effect by the simultaneous elimination of RARγ; this is demonstrated by the fact that Cyp26b1;RARγ double mutant embryos still show the proximalization of the PD molecular code, whereas chondrocyte differentiation is mostly rescued (Pennimpede et al., 2010).
Meis genes control spatial distribution of Hoxa13 expression via RA degradation
Meis misexpression in the distal limb bud has important consequences for limb development. The mechanisms by which Meis genes affect PD gene expression in the limb, however, had not been reported. Our current results show that MEIS factors affect RA degradation by controlling CYP26B1 expression in the limb bud (Fig. 2), and that RA degradation in turn controls Hoxa13 expression (Fig. 1). In fact, Hoxa13 repression upon Meis2 misexpression can be reversed by in utero treatment with an RA antagonist (Fig. 2). The fact that not all distal cells exposed to the RAA treatment recovered Hoxa13 expression is most likely explained by the low RAA dose used [5 mg/kg versus 10 mg/kg used in other studies (e.g. Wendling et al., 2000)]. We could not use higher doses because in our hands this compromised embryo viability, suggesting that the appropriate dose window (high enough to counteract Meis2a misexpression but low enough to allow embryo survival) is very narrow. We cannot, however, exclude another possible explanation of the incomplete rescue: that Meis genes affect Hoxa13 expression via multiple parallel pathways, RA being only one of them. Notably, the fact that the MEIS effect is cell-autonomous, despite acting through a diffusible signal (RA) is consistent with studies showing that RA degradation by CYP26 enzymes has a more determining role than RA diffusion on tissue patterning, very likely due to the different kinetics of the two processes (Hernandez et al., 2007; Probst et al., 2011; White et al., 2007).
Our results indicate that although the Meis1/2 expression domain never overlaps or abuts that of Hoxa13 (Mercader et al., 2009), the regulation of Cyp26b1 by Meis genes may contribute to the establishment of the PD distribution of RA, which would then affect the positioning of the Hoxa11-Hoxa13 transition even though this transition occurs when Meis expression in this region has already shut down.
Time is also necessary for Hoxa13 expression
To our knowledge, there are no studies showing premature expression of Hoxa13 upon physical, chemical or genetic manipulation of the limb bud. We confirmed this resilience to precocious expression by providing a distal signaling environment ahead of time in the early limb bud and observing that Hoxa13 could not be prematurely activated (Fig. 3). This lack of response was not due to problems with the treatments used, as they were able to expand or ectopically induce Hoxa13 expression once the endogenous domain already existed (Fig. 3B,D). It is noteworthy that when distal signals were provided the ectopic activation of Hoxa13 could take place quite proximally, although restricted to a posterior region (Fig. 3D). The posterior restriction of the ectopic activation might not be related to SHH signaling, as SHH was also added exogenously. We speculate that Hoxa13 activation may require additional factors present in the posterior mesenchyme or released from the posterior ectoderm, which has indeed been described as a signaling center influencing posterior mesenchymal expression (Nissim et al., 2007).
These results show that the early limb bud is not competent to activate Hoxa13 (and presumably the whole autopod program), and that time must elapse for the limb cells to become competent. Moreover, given that Hoxa13 can later be ectopically activated in the proximal region, the temporal competence seems to eventually apply to the whole limb bud, but the kinetics of the process is currently unknown.
Chromatin opening, not signal integration, controls the timing of Hoxa13 activation
Our data confirm that the timing mechanism does not rely on the integration of FGF signaling over time (Fig. 4), something that was previously suggested by the inability of excess FGF to activate Hoxa13 prematurely (Fig. 3) (Vargesson et al., 2001). Importantly, this does not mean that FGF signaling is not necessary for Hoxa13 expression during normal limb development. Indeed, we show that FGF signaling is needed to keep RA signaling away from the distal region, but that its role is merely permissive and not instructive: if RA signaling is artificially blocked from the distal region, the timing mechanism can proceed in the absence of FGF signaling (Fig. 4).
The alternative mechanism that we explored concerns chromatin state. It has been proposed that regulation of HoxD expression differs considerably between the limb [two transcriptional waves dependent on the interaction of Hox loci with distinct topological domains (Andrey et al., 2013; Montavon et al., 2011)] and the tail bud [strict progressive collinear activation dependent on progressive elimination of repressive histone marks (Soshnikova and Duboule, 2009)]. There is even a third scenario - based on in vitro experiments recapitulating rostrocaudal patterning of the motoneurons in the spinal cord - where the elimination of repressive histone marks, instead of being progressive, occurs in a rapid, domain-wide manner, even though the activation of Hox expression is progressive (Mazzoni et al., 2013). HoxA expression in the limb (i.e. progressive collinear activation, especially from Hoxa10 to Hoxa13) bears more similarities with the scenarios proposed in the tail bud or the spinal cord studies, and we therefore tested whether premature chromatin opening could trigger precocious Hoxa13 activation. One of the histone marks that correlates with transcriptional Hox activation is histone 3 lysine 27 acetylation (H3K27ac) (Soshnikova and Duboule, 2009) and we indeed found that if the histone acetylation/deacetylation balance is tilted towards acetylation by inhibiting HDAC activity in the early limb bud, Hoxa13 is prematurely activated (Fig. 5). Importantly, TSA applied systemically does not substantially alter normal development, except for a slight increase in the number of somites of mid-gestation mouse embryos (Nervi et al., 2001). In the limb, TSA does not cause a general upregulation of gene expression (Zhao et al., 2009a), and we found that it inhibits a number of distal limb markers, which suggests that the effect on Hoxa13 activation is not due to unspecific unleashing of normally repressed genes. Regarding the skeletal phenotype, the fact that the zeugopod and not the autopod is primarily affected, despite the distal location of the TSA bead, suggests that the premature activation of the autopod program leads to a reduction in the number of cells with a zeugopod expression profile crossing the differentiation front, and this eventually jeopardizes zeugopod size. It is noteworthy that a previous study using TSA on early limb buds claimed to find no effect on PD patterning (Towers et al., 2008), which is obviously at odds with our results. However, close examination of the results of that study reveals that the mildly affected specimens showed a preferential reduction of the zeugopod (Towers et al., 2008), which is consistent with our results and interpretation.
Previous studies of the regulation of Hox expression by chromatin state have focused on the molecules responsible for histone methylation/demethylation - mainly members of the Polycomb repressive complex 2 (PRC2) and the jumonji-domain containing demethylases - rather than on HDACs (Lan et al., 2007; Schorderet et al., 2013; Williamson et al., 2012). However, histone methylation and acetylation have been shown to be interdependent and to act as a single pathway in several tissues (Kleer et al., 2003; van der Vlag and Otte, 1999), justifying our approach. In fact, elimination of the PRC2 member Ezh2 from the early limb bud affects Hoxa13 expression (among other effects) in a way that could be interpreted as a precocious anterior expansion of the domain (Wyngaarden et al., 2011).
An interesting pending question is how is the timing mechanism triggered? A study in a culture model of rostro-caudal patterning of the spinal cord showed that sequential action of the patterning signals RA, FGF and Wnts is responsible for the progressive activation of Hox gene expression in a context of saltatory relief of histone repressive marks (Mazzoni et al., 2013). It was further speculated that the signals responsible for Hox10-Hox13 paralog activation could be FGFs and growth differentiation factor 11 (GDF11) (Mazzoni et al., 2013), and indeed GDF11 has been shown to ectopically activate Hoxd11 and Hoxd13 in the limb bud, but not Hoxa11 or Hoxa13 (Gamer et al., 2001), again underscoring the different behaviors of HoxA and HoxD genes in the limb. In this respect it is noteworthy that Hoxa13 is the only Hox gene that is always and exclusively expressed in the autopod precursors (Lu et al., 2008), and thus is expected to undergo different regulation than the more broadly-expressed HoxD genes. It remains possible, however, that the same signal postulated to trigger the switch between telomeric- and centromeric-interactive conformations for the HoxD loci, namely FGFs from the AER (Andrey et al., 2013), also triggers the activation of the HoxA timing mechanism, ensuring the coupling of both processes and hence robustness of limb development. If this were the case, it should be noted that only an initial pulse of FGF, enough to trigger the first PD subdivision, would be necessary for the timing mechanism to act, because our results show that after the Meis-Hoxa11 transition, continuous FGF signaling is no longer necessary as an instructive cue to maintain the process (Fig. 4).
A dual mechanism model of limb PD patterning
The results presented here, together with previous evidence (Cooper et al., 2011; Roselló-Diez et al., 2011), demonstrate the co-existence of two parallel mechanisms during limb PD patterning: one based on the signaling environment and a second based on chromatin regulation (Fig. 6). The limb bud is induced as a secondary axis that inherits the signaling milieu from the LPM, which is enriched in RA (Dollé et al., 2010). Therefore, when limb bud outgrowth starts, Meis genes are expressed, and the stylopod is specified (Fig. 6A). Subsequently, FGF signaling accumulates distally and RA is diluted in that region due to its growth away from the Raldh2-expressing flank, until the RA/FGF ratio drops below the first threshold, provoking distal Hoxa11 activation and Meis1/2 downregulation (Mercader et al., 2009) (Fig. 6A,B). The distal cells at this stage are therefore primed to become zeugopod, but their final fate will be established only once they cross the differentiation front (Tabin and Wolpert, 2007). Simultaneously, the elimination of MEIS activity from the distal limb cells allows CYP26B1 to start RA degradation in those cells, further decreasing RA levels until the permissive threshold for Hoxa13 activation is reached (Fig. 6C). Given that TSA can trigger premature Hoxa13 activation (Fig. 5) but that RAA cannot (Fig. 3), it follows that the permissive signaling environment is reached in the distal region before the chromatin-based mechanism triggers Hoxa13 expression (Fig. 6C). The underlying signal threshold map is therefore revealed only once the temporal constraint is released and both mechanisms converge to activate Hoxa13 (Fig. 6D). Once Hoxa13 expression starts, the shape of its expression domain therefore reflects the RA signal map in the limb bud. The initial posterior bias of the Hoxa13 domain (Mercader et al., 2009; Nelson et al., 1996; Yokouchi et al., 1991) could be due to at least two non-exclusive possibilities: (1) the fact that SHH, released from the posterior region, enhances CYP26B1-mediated RA clearance (Probst et al., 2011); and (2) transient repression of Hoxa13 by PRC2 in the anterior region of the limb bud, as Ezh2 elimination from the limb leads to anterior expansion of the early Hoxa13 expression domain (Wyngaarden et al., 2011). Finally, Hoxa13-mediated repression of Hoxa11 then establishes exclusive Hoxa13 expression in autopod precursors (Fig. 6D).
Zeugopod expansion requires delayed Hoxa13 activation
In the proposed dual model, the delay in Hoxa13 activation is essential, because it provides the time needed for accumulation of sufficient zeugopod precursors beyond the differentiation front to ensure proper zeugopod formation. If this delay is shortened, the model predicts that the main affected segment will be the zeugopod, as in fact happens with the TSA treatment (Fig. 5), setting the grounds for the interpretation of human intercalary limb defects.
The proposed model explains and integrates the main observations from classical embryological studies and those from recent genetic approaches, a goal that has remained elusive until now. In addition, the discovery of epigenetic regulation as a determinant of timed Hoxa13 activation, and its integration with signals, reveals a new mechanism by which chromatin regulation encodes a temporal delay essential for spatial patterning during embryonic development. This mechanism may be important during the regionalization of other growing embryonic primordia and explain the origin of certain congenital defects.
MATERIALS AND METHODS
AG1-X2 beads (Bio-Rad) were soaked in DMSO containing R116010 (0.5 mg/ml), TSA (0.5 mg/ml), SU5402 (2.5 mg/ml) or BMS493 (2.5 mg/ml). For simultaneous inhibition of FGF and RA signaling in grafts, beads were soaked in DMSO containing a mix of SU5402 (5 mg/ml) and BMS493 (4 mg/ml). Two beads were inserted per graft. Heparin-coated acrylic beads (Sigma) were soaked in PBS containing 0.1% BSA and 1 mg/ml FGF8 or 3 mg/ml SHH.
Embryos were staged according to Hamburger and Hamilton (Hamburger and Hamilton, 1951). In a PBS dish, distal tips of the desired width were cut with a microscalpel. Graft sites were prepared by making small rectangular wounds in the host tissue with an electrolytically sharpened tungsten needle, and the grafts were pipetted onto the wound and pushed in with a pair of blunt forceps. No staples or pins were used.
For the generation of R26loxP-STOP-loxP-Meis2a-IRES-eYFP (R26RM2) mice, all DNA constructs were assembled using standard digestion-ligation cloning methods. The Meis2a-coding sequence was PCR-amplified from 10.5-11.5 day post-coitum mouse embryo cDNA. This fragment and an IRES-eYFP cassette were cloned into pBigT (Srinivas et al., 2001) to generate pBigT-Meis2a-IRES-eYFP. The KpnI site in pROSA26PA (Srinivas et al., 2001) was changed to SwaI using a synthetic adaptor to produce pROSA26PAS. The Meis2a-IRES-eYFP fragment from pBigT-Meis2a-IRES-eYFP was PacI-AscI-cloned into pROSA26PAS to yield the final targeting vector. To generate R26RM2 mice, the targeting vector was linearized at the SwaI site and electroporated into ES cells (Torres, 1997). ES cell clones were screened by Southern blot. Chimaeras were generated by ES cell aggregation, and mice were subsequently genotyped by PCR (Soriano, 1999).
Timed matings were set up to generate embryos of the desired stages. The day of vaginal plug detection was considered embryonic day (E) 0.5. 4-Hydroxy-tamoxifen (4HT) (Sigma) was dissolved in corn oil (5 mg/ml) and injected intra-peritoneally (0.8 mg single dose, unless otherwise stated). For in utero treatment with RA antagonist, a solution of 50 mM BMS493 in ethanol was diluted in olive oil (1:6 v/v) and administered by oral gavage at two time points (E9.75 and 10.25) to pregnant mice at a dose of 5 mg/kg.
Cartilage staining, skeletal element measurements and statistics
Victoria Blue staining was performed as previously described (Carlson et al., 1986). Skeletal elements were measured with Adobe Photoshop by drawing straight lines along the middle of each skeletal element, breaking the line at the appropriate angle when necessary. The sum of the lengths of the different lines running through a particular element was then calculated. In this way, the analysis was not confounded by size changes due exclusively to distortion (bending).
Length differences between control and experimental skeletal elements were tested for statistical significance by one-way ANOVA followed by Tukey’s honest significant difference test to control for test multiplicity. Normality and homogeneity of variance were checked with the Kolomogorov-Smirnov test. Grubb’s test (α=0.01) did not detect any outlier in the data.
(E)-4-[2-[5,6-Dihydro-5,5-dimethyl-8-(2-phenylethynyl)naphthalene-2-yl] ethen-1-yl] benzoic acid (BMS493) and [1S, 2S)]-N-[4-[2-(dimethylamino)-1-(1H-imidazole-1-yl)propyl]-phenyl]-2-benzothiazolamine (R116010) were synthesized by InnoChemie GmbH (Würzburg, Germany). 3-[3-(2-Carboxyethyl)-4-methylpyrrol-2-methylidenyl]-2-indolinone (SU5402) was purchased from Merck KGaA (Darmstadt, Germany). Trichostatin A (TSA, pan-HDAC inhibitor) was purchased from Selleck Chemicals (Houston, TX, USA).
Recombinant mouse FGF8b and recombinant mouse SHH N-terminus were purchased from R&D Systems (Minneapolis, MN, USA).
MEIS2 isoforms were detected on cryosections as described previously (Mercader et al., 2005), except that we skipped the dehydration/rehydration steps prior to the cryoprotection.
In situ hybridization
Whole-mount in situ hybridization on mouse or chick embryos was performed as described previously (Wilkinson and Nieto, 1993). Proteinase K was diluted in PBS containing 0.1% Tween 20 (PBST) and incubations conducted as follows. Chick embryos: 20 mg/ml for HH20 (20 minutes) or HH22 (25 minutes); 30 mg/ml, 30 minutes for HH24. Mouse embryos: 12 mg/ml, 12 minutes for E10.5; 15 mg/ml, 15 minutes for E11.5.
In situ hybridization on paraffin sections was performed as described previously (Wilkinson and Nieto, 1993). Some sections were counterstained with nuclear Fast Red 0.005% before dehydration and mounting.
Note added in proof
A recent study has shown that although Hoxa genes display a 5′-3′ partitioning similar to their Hoxd counterpart (Woltering et al., 2014), most of the observed partitioning (including that of Hoxa13) is constitutive across different tissues and quite independent of transcriptional activity, suggesting that these contacts might rather act as ‘a priming mechanism for enhancer promoter interactions...by providing a stable framework to be complemented by tissue-specific factors’. It is tempting to speculate that the regulatory mechanisms we have described here for Hoxa13 transcription operate on top of that framework.
We thank J.M. González-Rosa for statistical advice; Sagrario Ortega, Brian Harfe and Susan Mackem for the Flpe, ShhGfpCre and HoxB6-CreER lines, respectively; Alexandra Joyner for generous access to reagents and equipment; and Simon Bartlett for editing. We also thank the reviewers for their suggestions to improve the manuscript.
M.T. conceived the project. C.G.A. performed MEIS immunohistochemistry and generated the R26RM2 targeting vector, G.G. performed the ES cell targeting, I.D. performed the in utero treatments with RAA, and A.R.-D. performed all other experiments. M.T. and A.R.-D. designed experiments, interpreted results and wrote the manuscript.
This work was supported by the Regional Government of Madrid [fellowship CPI/0050/2007 to A.R.-D.] and by the Spanish Ministry of Economy and Competitiveness [grants SAF2000-00160 and BFU2012-31086 to M.T.]. The CNIC is supported by the Ministry of Economy and Competitiveness and the Pro-CNIC Foundation.
The authors declare no competing financial interests.