Vertebrae display distinct morphological features at different levels of the body axis. Links between collinear Hox gene activation and the progressive mode of body axis elongation have provided a fascinating blueprint of the mechanisms for establishing these morphological identities. In this review, we first discuss the regulation and possible role of collinear Hox gene activation during body formation and then highlight the direct role of Hox genes in controlling cellular movements during gastrulation, therefore contributing to body formation. Additional related research aspects, such as imaging of chromatin regulation, roles of micro RNAs and evolutional findings are also discussed.
A characteristic feature of the vertebral column is its distinct morphology along the antero-posterior (AP) body axis, making it easy, for example, to distinguish between a cervical vertebra and that of the thoracic region by the presence of ribs. Another characteristic is its periodic structure that illustrates the repetition of vertebral segments along the body axis. The vertebrae and their associated muscles derive from metameric structures of the embryonic mesoderm, called somites. This metameric characteristic of the trunk is established by somitogenesis during which somites form sequentially in a rhythmic fashion from the presomitic mesoderm (PSM); this process of somitogenesis involves a molecular oscillator called the segmentation clock (Saga & Takeda 2001; Bessho & Kageyama 2003; Pourquie 2003; Aulehla & Herrmann 2004; Dubrulle & Pourquié 2004; Aulehla & Pourquié 2006). The PSM, as well as other mesoderm derivatives is formed sequentially through gastrulation and eventually becomes flanked by the neural tube.
Gastrulation is a key embryonic process that results in the formation of the three definitive germ layers and is a major driving force behind axis elongation that establishes the embryonic body plan. Vertebrate embryos develop in a similar progressive fashion in which anterior structures are formed first, and then more posterior structures are progressively added by gastrulation following the primitive streak and tail bud regression. Therefore, the position of these structures along the AP axis in an embryo directly reflects the timing of their production. This body plan is highly invariable among species, suggesting an evolutional constraint in which the temporal control of tissue formation and spatial regionalization along the AP axis must be highly coordinated during embryonic development. Although morphologically very similar, somites will differentiate into morphologically distinct vertebrae depending on their axial level which ultimately defines the axial formula (e.g. the occipital, cervical, thoracic, lumbar, sacral and caudal spine).
This positional identity of somites is specified largely by their unique combinatorial expression of Hox genes (Kessel & Gruss 1991). Mammals have 39 Hox genes that are organized into four clusters. In each cluster, the genes are arranged on the chromosome in a sequence that reflects their order of expression during embryogenesis and their expression domains along the AP axis (Fig. 1). The former correlation has been described as ‘temporal collinearity’ while the latter has been described as ‘spatial collinearity’ (Gaunt 1988; Dolle et al. 1989; Graham et al. 1989).
The vertebrate body axis progressively extends in a posterior fashion when new tissues are added from the gastrulation site at its posterior end. Hox clusters initiate their activity in gastrulating cells in a collinear manner in which 3′Hox genes are expressed first, then later, more 5′Hox genes become progressively activated during axis elongation. The collinear activation of Hox genes in gastrulating cells coupled with the mode of body axis elongation results in cells being produced earlier to express more 3′Hox genes and being located more anteriorly. This correlation led to the idea that the spatial expression of Hox genes along the AP axis is merely a readout of the progressive mode of axis extension in vertebrates. A number of extensive molecular analyses of Hox gene expression suggest a variety of complex strategies during embryogenesis, such as chromatin remodeling, long- and short-range promoter regulations and tissue-lineage specific regulations (Kmita & Duboule 2003; Deschamps & van Nes 2005).
This review focuses on the regulation and possible function of Hox genes during paraxial mesoderm formation. We primarily discuss molecular insights into the temporal and spatial collinearities of Hox genes as a conceptual framework for translation of embryonic time in space. We further discuss a recent observation of Hox gene function that links cell movement in gastrulation to AP patterning (Iimura & Pourquié 2006). The collinearities of Hox genes are also discussed in light of other findings, including imaging of chromatin regulation, micro RNA (miRNA)-based regulations and the evolutionary origin of this Hox system.
Qualitative and quantitative aspects for Hox function, Hox code and posterior prevalence
Drosophila has a single set of eight Hox genes that are grouped in the HOM complex. Numerous studies of the role of Hox proteins in fly development provide a clear view of their function. In the fly, Hox proteins control the identity of various embryonic domains along the AP axis by regulating distinct downstream targets. In vertebrates, however, evidence for quantitative as well as for qualitative aspects of the function of Hox proteins has been provided. The ancestral Hox gene cluster has been repeatedly duplicated, resulting in, for example, seven sets of Hox genes in fish and four sets in birds and mammals. The duplications of the ancestral Hox cluster generated groups of paralogue genes that share high similarities in sequence, expression pattern and function. In fact, substitution of Hoxa3 or Hoxa1 by Hoxd3 or Hoxb1, respectively, by a knock-in strategy in the mouse shows that these paralogue genes carry out identical functions and also suggests a quantitative aspect of the function of these genes (Greer et al. 2000; Tvrdik & Capecchi 2006). It has been proposed that the subtle morphological differences among different vertebrae reflect slight quantitative modifications acting on the downstream targets rather than on a set of distinct targets. Combinations of Hox4 paralogue mutants show a dose-dependent increase in the number of affected vertebrae (Horan et al. 1995).
The idea that the most posterior genes expressed in a particular embryonic domain impose their dominant function on more anterior genes is described as the posterior prevalence or phenotypic suppression of Hox genes (Duboule 1991; Duboule & Morata 1994). In many cases, the posterior prevalence concept accounts for phenotypes of Hox mutants better than the Hox code concept in which a combinatorial distribution of Hox products leads to morphological diversity of an individual vertebra (Kessel & Gruss 1991). For example, in null mutants for all the paralogues in a group, phenotypes are restricted to the narrow window of the anterior expression domain of these paralogues (Horan et al. 1995; van den Akker et al. 2001; Wellik & Capecchi 2003); whereas, the posterior domain overlapped by more posterior Hox genes shows intact morphology, which must therefore be imposed by the posterior genes (Fig. 2A,B). According to the Hox code hypothesis, changes in the combinatorial outputs would impose phenotypes in the whole expression domain of these mutated genes (Fig. 2C,D).
A conserved feature of the vertebrate axis is the order of morphologically distinct regions along the AP axis level (e.g. cervical, thoracic, lumbar, sacral and caudal vertebrae). Yet, different numbers of segments contribute to various regions resulting in the variation of axial formulae between vertebrate species. These morphological transpositions are paralleled by corresponding transpositions of the Hox gene anterior expression boundaries, suggesting an evolutionally conserved role of Hox genes in the positioning of vertebral domains rather than in the specification of each segment (Burke et al. 1995). This observation suggests a conserved qualitative function of Hox paralogues on defining an embryonic area.
Taken together, the posterior prevalence is an important property in evolution and could account for the diversity of the vertebral formulae among vertebrate taxa since the anterior or posterior shift of a single or a set of Hox genes could result in transposition of vertebrae. Therefore, the strategy of the collinear distribution of Hox products along the body axis and its generating mechanisms could have been evolutionally selected as an essential constraint linked to the body plan along the AP axis. Interestingly, chick Hoxc8 early enhancer shows delayed expression of a reporter gene in transgenic mice, resulting in a more posterior expression than the endogenous mouse Hoxc8 at a level that corresponds to the Hoxc8 expression in the chick embryo (Belting et al. 1998). Comparison of the early enhancer sequences between these two species shows only a few nucleotide differences, suggesting that limited changes in cis-regulatory sequences during evolution could be involved in the diversity of the axis formula.
Molecular mechanisms underlying the posterior prevalence remain unclear. Given that Hox gene expression largely overlaps along the body axis, it is unlikely that posterior genes repress transcription of anterior genes. miRNA-based post-transcriptional regulation could be involved in this mechanism (Ronshaugen et al. 2005). Demonstration of the posterior prevalence from co-overexpression of two Hox gene constructs from chick β-actin promoter in the chick embryo supports the idea that the suppression of the anterior gene by the posterior gene occurs at the protein level (Iimura & Pourquié 2006). This issue will be discussed later in this review.
Initiation of Hox gene expression and the potential role of signaling molecules
Temporal collinearity of Hox gene expression is first observed in the posterior streak region at the boundary between embryonic and extra embryonic tissues in mouse (Deschamps et al. 1999). For each gene, expression spreads anteriorly to reach the anterior primitive streak or node region where paraxial and axial mesoderm, respectively, are formed. This progressive mode of expression has been described as anterior propagation, forward spreading or rostral expansion (Gaunt 1988; Deschamps et al. 1999; Deschamps & van Nes 2005). Detailed fate mapping of the epiblast and careful observation of Hoxb gene initiation in the chick embryo demonstrated that Hoxb gene transcription is sequentially initiated in the epiblast at the level of the middle-to-posterior streak, which gives rise to mainly lateral plate and extraembryonic mesoderm (Iimura & Pourquié 2006). Subsequently, the expression domain spreads anteriorly through the paraxial mesoderm territory to reach the level of Hensen's node and the posterior neural plate. A similar initiation pattern is observed in the Xenopus marginal zone (Wacker et al. 2004).
Initial activation of Hox gene transcription and the early anterior propagation event of Hox gene expression could be regulated by molecular events involved in gastrulation, including the formation and caudal regression of the primitive streak. It has been proposed that Wnt signaling regulates the anterior progression of Hox genes since Wnt signaling regulates the formation of the primitive streak (Ikeya & Takada 2001; Forlani et al. 2003). Fgf signaling is also essential for mesoderm formation through the primitive streak and caudal regression of the primitive streak, suggesting that it could play a regulatory role in the initial expression of Hox genes as well (Ciruna & Rossant 2001; Dubrulle et al. 2001). It has also been suggested that retinoic acid (RA) could play a role in the initial activation of Hox genes because RA treatment of mouse embryos causes an anterior shift of 3′ Hox gene expression domains in somites. (Kessel & Gruss 1991; Roelen et al. 2002). Retinoic Acid, Fgf and Gdf11 have been shown to regulate anterior, intermediate and posterior Hox genes, respectively, in the developing neural tube (Liu et al. 2001; Bel-Vialar et al. 2002) but their role in the initial activation of Hox genes in the mesoderm remains to be clarified (Gamer et al. 1999; McPherron et al. 1999). As observed in the precursors of the hindbrain and spinal cord (Nordstrom et al. 2006), combinatorial roles of these signaling molecules may be involved in Hox gene initiation in the PSM precursors in the epiblast.
Genome structure and collinear activation
A functional link between the tight collinear arrangement of Hox genes in the genome and the temporal collinear activation during embryogenesis has been postulated for vertebrates (Kmita & Duboule 2003). In contrast, the Drosophila HOM complex is split into two complexes, Antenapedia and Bithorax, and spans more than 1 Mb in length (Duboule 1994). Whereas expression of these genes respects the spatial collinearity in the fly, no temporal collinearity of their activation has been observed. In vertebrates, genes are collinearly arranged in intact Hox clusters of approximately 100 kb in length and are expressed in a temporal collinear fashion. Human genome sequencing demonstrates the extremely low density of interspersed repeats (as Alu sequence) in the four HOX clusters, suggesting that large-scale cis-regulatory elements in each cluster are a strong selective constraint that cannot tolerate being interrupted by insertions (Lander et al. 2001).
It has been proposed that this characteristic feature of vertebrate Hox clusters is associated with the progressive activation of Hox genes from one end of the cluster to the other (Kmita & Duboule 2003). Relocation of Hoxd9/LacZ and Hoxd11/LacZ transgenes at the 5′ extremity of the Hoxd complex (between Hoxd13 and Evx2, see Fig. 1A) caused a delayed activation of these transgenes, behaving like Hoxd13 at the early stages when compared to the same transgenes that are randomly integrated, suggesting a repressive mechanism due to the location within the cluster (van der Hoeven et al. 1996). Furthermore, a conditional fusion of the 5′ exon of Hoxd13 with the 3′ exon of Hoxd12 was expressed like Hoxd11 with a more anterior expression domain than expected for either (Kondo et al. 1998). This observation argues for the existence of a regulatory element between Hoxd13 and Hoxd12 that contributes to the time of activation of these genes. Successive insertion of a Hoxd9/LacZ transgene upstream of the Hoxd cluster followed by a series of deletions identified a global regulatory element in the 5′ end of the Hoxd complex. This element is essential for the early collinear activation of the cluster (Kondo & Duboule 1999). Together, these observations suggest a silencing mechanism originating at the 5′ end of the cluster, which sequentially makes genes available for transcription in a 3′ to 5′ order. However, relocation of Hoxb1/LacZ transgene into a position upstream to the Hoxd cluster hampered this idea since the transgene expression was not suppressed in the mesoderm, and even more unexpectedly, induced an ectopic and transient activation of the neighboring Hoxd13 (Kmita et al. 2000). This observation indicates that the timing of activation of a Hox gene in the cluster does not exclusively depend on its position, but also suggests that further regulation is necessary to set up the spatio-temporal pattern.
A basic mechanism could be that repressors or silencing factors are released from the 3′ to 5′ end through a transition in chromatin status. In fact, the zinc finger repressor protein, Plzf, controls the expression of posterior Hoxd genes through chromatin remodeling (Barna et al. 2002). Also, ectopic expression of Hoxb1 and Hoxb2 in response to RA depends on whether they are within or outside of the cluster (Roelen et al. 2002). A randomly integrated Hoxb1/LacZ transgene that contains retinoic acid response elements (RARE) faithfully mimics the endogenous spatial expression pattern of Hoxb1; however, unlike the endogenous locus, it does not cause immature activation by short RA exposure. These observations indicate that the initial transcriptional availability of Hox genes depends on their location in the cluster and on specific regulatory elements.
Distinct interactions between distal cis-control by a global enhancer located outside of a cluster and local cis-regulatory elements adjacent to each Hox gene further ensure the collinear pattern of Hox genes (Kmita & Duboule 2003; Deschamps & van Nes 2005). These interactions probably involve further tissue-specific regulations, such as Hox protein interregulation that has been implicated in the hindbrain development (Gavalas et al. 2003; Serpente et al. 2005). In the PSM, the spatial Hox pattern is coupled with segmentation machinery (Dubrulle et al. 2001; Zakany et al. 2001) which will be further discussed in the following section. It remains to be determined how much of the initial collinear activation of Hox genes accounts for the establishment of the collinear distribution of Hox proteins along the body axis.
Relevance of early Hox gene expression for morphological identity of somite derivatives
Chick transplantations and mouse genetic experiments have led to the idea that Hox gene expression is correlated to the acquisition of future somite identity as early as the unsegmented PSM stage. After heterochronic transplantations of the PSM, grafted cells still maintain their original Hox gene expression and tend to differentiate according to their original identity even when placed at a distinct axial level in the host (Kieny et al. 1972; Jacob et al. 1975; Nowicki & Burke 2000). The relevance of the precise temporal activation of Hox genes is also supported by a mutation of Hoxc8 early enhancer in the mouse. This mutation, which causes an initial delay of Hox gene expression, affects skeletal patterning by largely phenocopying the Hoxc8 null mutant even if spatial somitic expression appeared to be recovered at later stages (Juan & Ruddle 2003). Other mutations delaying Hox gene expression caused homeotic transformation of vertebrae (van der Hoeven et al. 1996; Kondo & Duboule 1999) as observed in the Hoxc8 mutant. Therefore, activation of Hox genes at an early stage of paraxial mesoderm development is crucial in establishing regional identity.
It has been proposed that the spatial regulation of Hoxd genes in the anterior PSM is controlled by the segmentation machinery (Zakany et al. 2001). Replacing the entire Hoxd cluster with a LacZ reporter showed periodic stripes of gene expression in the anterior PSM in concert with segmentation, suggesting that a global enhancer outside of the cluster is responsible for the segmental expression in the PSM. This regulatory expression is under the influence of the Notch pathway. In fact, loss- or gain-of-function mutations in genes of the Notch pathway affect Hox gene expression in the PSM and subsequently, skeletal patterning (Cordes et al. 2004). Furthermore, overexpression of Hox genes driven by a PSM-specific promoter shows a severe effect on skeletal patterning; whereas, overexpression from a promoter driving expression in somites does not elicit such phenotypes (Carapuco et al. 2005).
Active role of Hox genes on mesoderm formation
We observed that collinear activation of Hoxb genes regulates the timing of mesoderm formation through controlling the flux of epiblast cells to the primitive streak (Iimura & Pourquié 2006). The initial activation of the Hoxb cluster in paraxial mesoderm precursors was carefully investigated in the chick embryo. Hoxb genes, from Hoxb1 to Hoxb9, are sequentially activated in the somitic territory of the epiblast. Interestingly, activation of each Hox gene was first observed in the epiblast layer showing a salt-and-pepper pattern. Expression then spread through the epiblast layer and eventually expanded to the primitive streak cells and ingressing mesoderm.
These expression dynamics follow the gastrulation movements of cells from the epiblast to the mesoderm, suggesting that Hox genes could play a role in mesoderm formation through gastrulation. This hypothesis was tested by electroporation of Hox-expression constructs containing a fluorescent reporter. Overexpression of Hoxb genes in chick embryos in the presumptive territory of somites in the epiblast at the gastrula stage resulted in collinear phenotypes in terms of AP distribution of Hox-expressing cells. Cells overexpressing more anterior Hox genes contributed to more anterior portions of the somitic columns; whereas, cells overexpressing more posterior Hox genes contributed to more caudal domains. When posterior Hox genes (e.g. Hoxb7 or Hoxb9) were electroporated, the Hox-expressing cells stayed in the epiblast longer than cells expressing either control GFP or anterior Hox genes (e.g. Hoxb1 and Hoxb4) (Fig. 3). A deletion mutant of the third helix of Hoxb9 homeodomain did not show this delayed effect on epiblast ingression, suggesting that DNA binding is essential for this function. The posterior prevalence effect was directly confirmed in this context by coelectroporation of two different Hox genes marked by distinct fluorescent proteins. Moreover, using different promoter activities to drive Hox genes, qualitative rather than quantitative effects of each Hox gene were demonstrated.
Thus, these studies demonstrate that somite precursor cells in the epiblast activate Hox genes in a temporal collinear manner. In the initial activation phase of each Hox gene, only a few cells activate a given Hox gene so that these cells acquire slightly different cell properties based on the posterior prevalence effect and thus, remain in the epiblast longer than others. Such a cell-sorting type of selection operated by the sequential activation of Hox genes controls the timing of cell ingression to the primitive streak, therefore regulating the AP distribution of mesodermal cells (Fig. 4). This function of Hox genes on mesodermal cell distribution along the AP axis conjugated with their temporal collinear activation most likely provides initial information in establishing the spatial collinearity, and could provide a framework to understand the diversification of the axis formula among vertebrate species.
Visualization of chromatin dynamics
The temporal and spatial expression of Hox genes as discussed above has created particular interest in the role of chromatin dynamics in the early collinear activation of Hox genes. In fact, the polycomb group (pcG) and trithorax group (trxG) proteins, respectively, play an important role in silencing and activating the transcription of Hox genes during both fly and mouse embryogenesis (Ringrose & Paro 2004). The relationship between chromatin modification and Hoxd4 expression was observed by chromatin immunoprecipitation at the Hoxd4 locus in both RA-induced neuronal differentiation of P19 cells and mouse embryonic tissue (Rastegar et al. 2004). In both systems, histone modifications typical of active chromatin occur first at the 3′ region of Hoxd4 and sequentially distribute in a more 5′ direction. Recruitment of transcription factors, such as CREB binding factor (CBP) and RNA polymerase II, spreads toward the same direction. This directed chromatin activation mirrors the collinear activation of the Hox cluster. Furthermore, the chromatin activation and the recruitment of the transcriptional factors are observed in posterior embryonic tissue that expresses Hoxd4, but not in anterior tissue in which Hoxd4 is not transcribed. This observation indicates that changes in the chromatin status of a Hox gene are concomitant with its transcriptional regulation.
The sequential activation of genes along the Hoxb cluster was imaged in mouse embryonic stem (ES) cells (Chambeyron & Bickmore 2004) and in early mouse embryos (Chambeyron et al. 2005) by fluorescent in situ hybridization. First, chromatin modifications were analyzed across the Hoxb locus during RA-induced ES cell differentiation. Markers of active transcription in chromatin, such as lysine 9 acetylation and lysine 4 dimethylation of histone H3, were increased at both Hoxb1 and Hoxb9 loci at an early time point when only Hoxb1 was expressed. Thus, these histone modifications are not tightly linked to transcription. Furthermore, short exposure of ES cells to RA at a time when only Hoxb1 is activated is sufficient to induce a general decondensation of Hoxb locus and selectively induce Hoxb1 looping out of its chromosome territory (CT), while Hoxb9 is neither expressed nor looped out. Longer RA incubation, however, induces both Hoxb1 and Hoxb9 looping out of the CT. In gastrulating embryos, looping out of Hoxb1 from the CT was observed where Hoxb1 was first expressed. In the tail bud of an E9.5 embryo where both genes are expressed, both Hoxb1 and Hoxb9 looped out of the CT. These observations demonstrate that chromatin decondensation and successive looping out of the CT is tightly concomitant with gene transcription in both in vitro and in vivo contexts. Although a functional link between the looping out of genes and transcription remains to be established, extrusion of a locus from the CT probably represents a poised state for transcription. Mechanisms localizing a gene to a subnuclear position have been associated with structures that facilitate transcriptional control. For instance, direct interaction of PcG proteins and a spliceosomal protein suggest a tight association between nuclear structure and transcriptional machinery (Isono et al. 2005).
Post-transcriptional regulation by miRNAs
Another interesting aspect of the regulation of Hox expression is the miRNA-mediated, post-transcriptional regulation (Chopra & Mishra 2006). Three miRNAs, miR-196a-1, miR-196a-2 and miR-196b, can direct cleavage of Hoxb8 miRNA through their complementary sequence to this gene in vitro (Yekta et al. 2004). Hoxb, Hoxc and Hoxa clusters, respectively, encode these miRNAs between Hox9 and Hox10 genes at a very similar position to mir-iab4 in the Drosophila HOM complex that is located between abdA and abdB. These miRNAs are extensively conserved among vertebrates and potentially bind to the miRNAs of Hoxb8, Hoxc8, Hoxd8 and Hoxa7. This system was shown to be at work by targeting Hoxb8 3′UTR during posterior limb bud patterning, thus providing a safeguard mechanism against inappropriate Hox activity potentially involved in RA-induced Shh expression (Hornstein et al. 2005). Another miRNA, miR-181, targets Hoxa11 in the context of muscle cell differentiation (Naguibneva et al. 2006). Based on the collinear activation model, this miRNA locus will be activated later than its target; thus, the miRNA-mediated cleavage could be involved in the posterior prevalence (Ronshaugen et al. 2005). Functional involvement of miRNAs in Hox gene regulation in vertebrate body development remains to be demonstrated by genetic loss- and gain-of-function experiments.
Evolutional aspects of the collinear Hox system
The ancestor of chordates is believed to be a marine, soft-bodied, worm-like animal with a single Hox cluster (Garcia-Fernandez 2005; Ikuta & Saiga 2005; Lemons & McGinnis 2006). However, in urochordates, the collinear arrangement of the Hox cluster in the genome is lost to a large degree. The urochordate Oikopleura diocia shows an extreme divergence from the conventional collinear arrangement of Hox genes (Seo et al. 2004). This urochordate has a single set of Hox genes that is roughly expressed in a spatial manner along the AP axis of a developing embryo, but its Hox genes are not clustered in the genome and are split into distinct loci. Another urochordate, Ciona intestinalis, has a partial breakdown of its Hox gene arrangement, yet shows some degree of spatial collinearity along the AP axis (Ikuta et al. 2004). This loss of Hox clustering in the genome could be related to the loss of temporal collinearity. After fertilization, very early expression of anterior and posterior Hox genes occurs in these urochordates as well as in Drosophila and C. elegans, all of which have split Hox genomes. These observations suggest that local and short-range cis-regulatory elements might be enough to establish the spatial pattern in these species. Amphioxus, a cephalochordate, has a single Hox cluster of 14 collinear genes (Minguillon et al. 2005) which suggests that Hox gene clustering and its collinear activation are present in cephalochordates and vertebrates as suggested in their common ancestral origin. Conserved mechanisms of the global and long-range cis-regulation among these species or among the clusters in a species remain to be explored.
Since the Hox code model for vertebrae morphogenesis was proposed in 1991 (Kessel & Gruss 1991), a large number of data have provided mechanistic insights into how the spatial collinear pattern is established and how regional identity is imposed during paraxial mesoderm development. However, we still experience difficulty in accommodating these observations within a well-defined frame. Stepwise approaches using various experimental models will be required to dissect the molecular regulation and function of Hox genes and to provide a better understanding of the fascinating regulation of these genes and of their role in establishing the animal body plan.
The current work was supported by the Stowers Institute for Medical Research and NIH grant R01 HD043158 to OP. We thank Pourquié lab members, Alexander Aulehla, Bertrand Benazeraf and Cheng-Wen Su for critical reading of the manuscript and Danny Stark in the Imaging Center of the Stowers Institute for Medical Research for his expertise in imaging. Finally, we thank Silvia Esteban for artwork and Joanne Chatfield for editorial improvements.