Msx1 is expressed in retina endothelial cells at artery branching sites

Angiogenesis is the process by which pre-existent vessels rearrange in order to give rise to new vascular beds (Jain, 2003). The mouse retina is one of the best characterized models for in vivo vessel formation as it displays all of the morphological hallmarks of angiogenesis (i.e. sprouting, branching, fusion, remodelling and maturation) after birth (Fruttiger, 2002, 2007). In contrast to human, murine newborns do not possess a developed retinal vascular plexus (Gyllensten and Hellstrom, 1954). In these, the superficial retinal vascular plexus forms during the first week after birth by radial outgrowth of vessels from the optic nerve entry point to the periphery (Fruttiger, 2007; Stahl et al., 2010). These superficial vessels reach the retinal edges at approximately post-natal day (P) 9 and from this stage onwards, vertical sprouting forms the deep and then intermediate vascular beds, that reach the retinal periphery at approximately P12 and P15, respectively. The three retina vascular beds become fully mature and interconnected at the end of the third postnatal week (Stahl et al., 2010). At this stage, the retinal superficial vasculature is composed of six major arteries and veins that form primary branches into arterioles and venules. Arterioles and venules branch into a capillary network that is in direct contact with deep and intermediate layers. This network is covered with pericytes, providing retina with the densest coverage of pericytes in the whole vasculature (Shepro and Morel, 1993). The choroidal vascular system, which resides between the retina and the sclera, is fully developed before birth and supplies oxygen and nutrients to the avascular retina (Campochiaro, 2000). At early stages, the blood supply of the eye is also provided by the hyaloid vasculature that originates from the central hyaloid artery in the optic nerve and extends through the primitive vitreous toward the anterior segment (Saint-Geniez and D'Amore, 2004). The hyaloid vasculature regresses when ocular development proceeds in order to leave a transparent visual axis (Mitchell and Gingras, 1998; Brown et al., 2005).

Blood vessels are mainly composed of an inner endothelial cell (EC) layer externally covered by mural cells, namely vascular smooth muscle cells (VSMCs) in arteries and veins, and pericytes in capillaries (Jain, 2003). In the mature vascular plexus, mural cells contribute to endothelial tube stabilization, maintenance of vascular permeability and regulation of the blood flow. In small arterioles and capillaries, most of these activities are performed by pericytes (Orlidge and D'Amore, 1987; Sato and Rifkin, 1989; Diaz-Flores et al., 2009; Armulik et al., 2011). In the central nervous system, not all the pericytes contain the muscle-specific actin isoform, only those in the pre- and postcapillary regions (i.e. on arterioles and venules) (Nehls and Drenckhahn, 1991; Rucker et al., 2000). This led to the proposition that blood flow is regulated mainly at the level of precapillary arterioles (Anderson and McIntosh, 1967; Baez, 1977). However, time-lapse and functional analyses have demonstrated that capillaries are capable of regulating blood flow along their whole length (Metea and Newman, 2006; Peppiatt et al., 2006). In the eye, vasoconstriction and vasodilatation via the pericytes are evoked by light-induced stimulation of perivascular astrocytes (Metea and Newman, 2006).

Msx genes encode homeodomain transcription factors and are essential for normal craniofacial and limb development, nervous system patterning, eye formation, as assessed by phenotypic abnormalities in knock-out mice (Bach et al., 2003; Wu et al., 2003; Ishii et al., 2005; Lallemand et al., 2005, 2009). In the cardiovascular system, Msx genes have previously been shown to be important for outflow tract patterning (Ishii et al., 2005; Chen et al., 2007), head vessel maturation (Lopes et al., 2011) and adult vessel calcification (Towler et al., 2006). In adult mice, Msx2^lacZ is mainly expressed in a subset of VSMCs in peripheral arteries and veins (brachial, femoral and caudal), as well as a few ECs of the aorta. Mxs1^lacZ is expressed to a lesser extent by VSMCs of peripheral arterial trunks, but is highly expressed in mural cells of arterioles and capillaries that irrigate the tissues (Goupille et al., 2008).

In this paper, we observe that Msx2^lacZ expression is not detectable in eye vessels, and describe Msx1 expression patterns in the retina and choroid vasculature. In the choroid, Msx1^lacZ is expressed in a broad dynamic pattern, initially in the endothelium. Expression is observed later on in VSMCs, similar to Msx1 general behaviour in the peripheral vasculature. In contrast, in the retina, Msx1^lacZ expression is restricted to ECs of the arteriolar branching points in the superficial vascular network. At these branch points, we further characterised a sub-population of pericytes, which express high levels of NG2, α-SMA and desmin. We thus demonstrate specific properties of cells constituting these branching structures, not only in the mural layer but also in the endothelial layer.

Outgrowth of vessels in the retina occurs postnatally from the central optical disc. This results in the formation of a stereotyped vascular network (Fruttiger, 2007). Analysis of Msx1^lacZ mice at P0 revealed that the primitive plexus was formed of ECs and did neither express smooth muscle actin, nor Msx1 (Fig. 1A,A′, arrowheads). The hyaloid vasculature, which is present during prenatal development and regresses shortly after birth, expressed strong levels of Msx1^lacZ as revealed by Xgal staining, and was covered with α-SMA-expressing mural cells (Fig. 1A,A′, arrows). Considering the morphology of the β-galactosidase (β-gal)-positive nuclei and their position relative to isolectin B4 (Ib4) and α-SMA domains, Msx1-expressing cells were likely ECs. At P3, retinal arteries have started to form from the primitive plexus. These arteries, which were part of the newly-formed vascular plexus (Fig. 1B′; data not shown), were covered with α-SMA-positive mural cells. In these arteries, Msx1 expression was detected in cells with elongated nuclei, suggesting they belonged to the endothelium (Fig. 1B,B′, arrows). At this stage, no Msx1 expression could be detected in less mature vessels not covered by mural cells (Fig. 1B,B′, arrowheads). At P10, the superficial vessels have reached the peripheral edge of the retina, the capillary bed has begun to develop into the deeper layers and circulation has started (Brown et al., 2005; Fruttiger, 2007). Strikingly, at this stage Msx1^lacZ expression became restricted to primary branching points in arterioles (Fig. 1C,C′, arrows). Restricted expression was not yet completed by P12, since a few cells were still expressing Msx1 along the length of arteries (Fig. 2A, arrowheads), but was achieved at P14 (Fig. 1D,D′). In most cases, more than one nucleus was labelled (Fig. 1F, Fig. 3D-F, Fig. 4A,C). At the branching points, we further observed strong Ib4 concentration that co-localized with Msx1-expressing cells (Fig. 1C′-F′, arrows). Ib4 is known to bind the basement membranes (specifically, the versican protein) and thus can label not only ECs but also microglial cells and probably other cell types including some vascular mural cells. On trypsin-digested retinas, Ib4 was unambiguously associated with vessels (not shown). Maximum Msx1 expression occurred between P14 and P23 (Fig. 1D–E), but expression was maintained at high levels up to P480 (Fig. 1F) and probably lasts over the lifetime of the animal. From P14, we observed that Msx1 also labelled secondary branches in the arteriolar network (Fig. 1F, Fig. 4D). Noticeably, at all stages, Msx1 expression was observed only in rather mature arteries that were covered by a SMA-positive mural cell coating suggesting that Msx1 expression in the endothelium requires EC-VSMC interactions. Msx1-labelled branching sites were quantitated throughout development (Fig. 2). There was a large and steady increase between P9 and P23 and thereafter the number plateaued. Therefore, restriction of the expression to branching sites lagged behind the formation of the vascular superficial bed, for which arteries reach the retinal periphery around P8 (Fruttiger, 2007). This was not achieved before the onset of blood flow, which in the mouse retina takes place between P3 and P4 (Brown et al., 2005). From P23, we observed Msx1 expression at all primary, and some secondary, branching points along the main retinal arteries.

Arteries and veins were identified in flat mounts by virtue of their distinct morphologies and differential coating by mural cells. We did not observe Msx1^lacZ labelling of venules in the retina (data not shown), as in any other veins (Goupille et al., 2008). Only the superficial vascular network was labelled. This is in keeping with the fact that deeper beds are primarily formed by sprouting from veins (Fruttiger, 2007). Xgal staining reflected the specific expression of Msx1 in retinal vessels since no staining was observed in normal mice (data not shown). In contrast to what was observed in peripheral arteries (Goupille et al., 2008), Msx2^lacZ could not be detected in hyaloid, retina or choroid vessels at any stage of their development (data not shown).

Xgal staining is useful for a broad analysis of expression patterns. However, it does not provide sufficiently high resolution for precise co-localization studies using optical microscopy. In order to identify in which cell type Msx1^lacZ is expressed, we used an anti β-gal antibody together with endothelial-specific (anti-CD31) and mural-specific (anti-α-SMA) antibodies, to perform confocal microscopy on flat mount retinas. This showed that the Msx1 reporter gene is expressed in ECs (Fig. 3B,E, arrowheads) and not in mural cells (Fig. 3A,D, arrows). Indeed, the nuclear β-gal signal was completely surrounded by CD31 protein and was observed in a luminal position relative to the mural marker α-SMA (Fig. 3C,F). As expected, β-gal colocalized with nuclear Hoechst staining, since the lacZ gene in this transgenic mouse is associated with a nuclear localisation sequence (not shown).

To get insight into the role of Msx1 at arteriolar branching points, we investigated possible correlations with accumulation of other proteins. We co-stained the retina from Msx1^lacZ animals for β-gal together with α-SMA (Fig. 4A,A′), desmin (Fig. 4B,B′) and NG2 (Fig. 4C,C′). In the rat retina, NG2 and desmin are expressed in immature mural cells around birth, but NG2 is expressed in pericytes in the adult and weakly in arteriole and vein VSMCs, whereas desmin remains at a high level in pericytes and VSMCs. On the contrary, α-SMA, which is also expressed in immature mural cells, becomes highly expressed in VSMCs and weakly in pericytes after birth (Hughes and Chan-Ling, 2004). According to our data, the primary branching structures were supported by a sub-population of mural cells that expressed α-SMA (Fig. 4A′), desmin (Fig. 4B′) and NG2 (Fig. 4C′). Of note, calponin is expressed primarily in VSMCs, although it has been described in pericytes of aged animals (Hughes and Chan-Ling, 2004; Hughes et al., 2006). Calponin was expressed in a fraction of branching structures, and always in cells on the distal side relative to the arteriole (not shown). Thus, at the branch sites, the sub-population of mural cells differs from others by its capacity to express simultaneously VSMC-associated proteins, such as α-SMA and calponin, together with pericyte markers like NG2 and desmin. Of note, α-SMA, desmin and calponin are proteins that participate in contraction, making this structure a candidate for specific contractile properties. Furthermore, according to Ib4 (Fig. 1C′–F′) and laminin (Fig. 4D′, arrows) expression levels, the density of the basement membrane at these branching points appeared high.

To further investigate the possible function of Msx genes in the retinal endothelium, we inactivated Msx1 and Msx2 specifically in ECs, taking advantage of the previously described Msx1^flox (Fu et al., 2007) and Msx2^flox (Bensoussan et al., 2008) alleles, and the Tie2-Cre transgene (Kisanuki et al., 2001). Only Msx1 is expressed in the retina; however Msx2 was also inactivated to avoid any compensation between the two genes (Ishii et al., 2005; Lopes et al., 2011).

Tie2-Cre has been shown to specifically induce loxP site recombination in the retina EC plexus (Ye et al., 2009; Weskamp et al., 2010; Sweet et al., 2011). However, no data has been reported referring to the efficiency of this transgene in ECs at branching site. To ascertain this point, we associated the Rosa^mT/mG reporter gene (Muzumdar et al., 2007), that expresses Tomato red before and Gfp after loxP recombination, with the Tie2-Cre transgene. GFP was revealed on dissected retinas using specific anti-GFP antibodies, together with antibodies against endothelial (CD31) or mural (α-SMA) markers. This demonstrated recombination in all ECs, including those residing at branching sites (supplementary material Fig. S1).

These data allowed us to perform the functional analysis. Dissected retinas from Msx1 ^flox/+ Msx2 ^flox/+ Tie2-Cre</emph> and Msx1 ^{flox/ flox} Msx2 ^{flox/ flox} Tie2-Cre mice were labelled for CD31, α-SMA and desmin, then analysed in flat mount by confocal microscopy. Msx1 and Msx2 inactivation in the endothelium did not result in major structural defects, neither at P7, when the vessels are intensely sprouting (Fig. 5A-F), nor in two months-old animals, when angiogenesis is achieved (Fig. 5G–L). At the two stages, CD31 staining (Fig. 5A,D, C,F and G–J) did not reveal either over- or under-branching in arteries, where Msx1^lacZ is expressed. Furthermore, considering the intensity of α-SMA (Fig. 5B,E and H,K) and desmin (Fig. 5I,L) signals, there was no conspicuous reduction in VSMC coverage or decrease in the expression of contractile proteins at branching sites. In the retina, desmin is strongly expressed in pericytes (Hughes and Chan-Ling, 2004), and we could not detect any difference in the number of desmin-positive cells in capillaries of mutant versus control embryos (Fig. 5I,L).

To verify whether this branch site-restricted expression pattern is more widespread in the eye vasculature, we investigated Msx1 expression in the choroid. In the mouse, choroidal vasculature is fully developed before birth and supplies oxygen and nutrients to the anterior region of the retina (Campochiaro, 2000). Similar to the retinal vasculature, it develops according to a stereotyped pattern, constituted of major arteries that grow from the optic nerve entry point toward the periphery. Msx1^lacZ expression was detectable as early E16.5 and became conspicuous at E17.5. At these stages, expression took place in arterioles covered with α-SMA-expressing mural cells and also in mural cell-free capillaries (data not shown). At P0, Msx1^lacZ was broadly expressed in choroidal vessels, at a high level (Fig. 6A,A′). On flat mounts, expression was observed in internal, longitudinal cells, which matched Ib4 labelling, suggesting it took place in ECs. Similar to prenatal stages, expression was observed in mural cell-covered (Fig. 6A,A′, arrows) as well as mural cell-free (Fig. 6A,A′, arrowheads) arteries. Surprisingly, at P14 and later, Msx1^lacZ expression was detected in cells resembling mural cells, since the β-gal-positive nuclei exhibited a rounder shape and a more external position (Fig. 6B,B′). Furthermore, expression was now restricted to mural cells-covered arteries. We analysed Msx1^lacZ expression until P150 and the location and intensity of Xgal staining did not change with age (Fig. 6C,C′).

To confirm the shift in expression of Msx1^lacZ from endothelial to mural cells, we performed a confocal analysis of choroid transverse sections. At P0, the nuclear β-gal protein was observed in most CD31-positive cells of the major arterioles (Fig. 7A, arrows). Clearly, α-SMA-positive cells did not accumulate β-gal protein (Fig. 7B, arrowheads). At P21 the localization of β-gal had changed dramatically: CD31-surrounded nuclei appeared completely devoid of β-gal protein (Fig. 7C, arrows) whereas α-SMA-positive cells expressed high levels of this protein (Fig. 7D, arrowheads). This change in pattern was observed in the whole choroid vascular tree as shown in Fig. 6.

Studying the different structures that compose the retina and choroid vasculature may give insight into disease processes such as diabetic or hypertensive retinopathy, hypertensive choroidopathy and macular degeneration. Taking advantage of Msx1^lacZ and Msx2^lacZ knock-in mice, we have established that Msx1, but not Msx2, is expressed in the mouse retinal and choroidal vasculature. One of the most striking results we obtained is the specific expression of Msx1 in a cluster of ECs at primary, and sometimes secondary, arteriolar branching sites in the retina. This property might be more widespread in the peripheral vasculature. We previously reported that Msx1 expression is more intense in arterioles of the thigh muscle at branching sites (Goupille et al., 2008). Our data suggested that expression took place in VSMCs, but the situation should be revisited using confocal microscopy at these sites. The expression pattern we observe in retina raises questions about the function Msx1 may play and the mechanisms that might activate it at branching sites.

Msx1 is unlikely to be associated with formation or stabilization of retinal arteriole branches, since disruption of the Msx1 gene specifically in the endothelium does not lead to defects in branching frequency or branch stability. A function related to vessel physiology is therefore more likely. In any case, Msx1 may be responding to specific signalling molecules at these sites. The Notch pathway is of particular relevance to this issue. The initial pattern of Jag1 expression in the retina is different from that of Msx1, but at P15, Jag1-positive cells are concentrated at branch points in arteries, in both endothelial and mural cells, in a pattern strikingly similar to Msx1 (Hofmann and Luisa Iruela-Arispe, 2007). However, Jag1 null mice die by E11.5 as a result of a lack of vascular remodelling (Xue et al., 1999), a phenotype very different from the one observed in Msx1 single or Msx1; Msx2 double null mutant (Satokata et al., 2000; Lallemand et al., 2005). Furthermore, Notch signalling plays a major role in controlling sprouting and at P6, in endothelium-specific Jag1 mutants, the density of newly formed vessels at the periphery of the retina is clearly reduced (Benedito et al., 2009). This phenotype was not observed in the endothelium-specific mutation of Msx1 (Fig. 5C,F). Altogether, these considerations imply that, if Msx1 is involved in the Notch pathway, it relays only a fraction of Notch activity. Tie1, an endothelial-specific receptor of the Tie receptor tyrosine kinase family with unknown ligand, is also expressed in the retinal vessels, and further concentrated at branching sites (Porat et al., 2004). Other signalling pathways, not yet characterized in the retinal vasculature, may play a role in Msx1 expression.

Conspicuously, branching sites in retinal arterioles form specific structures, which have been designated as arteriolar annuli (Henkind and De Oliveira, 1968; Simoens et al., 1992). These are characterized by hypercellularity and specific or enhanced expression of a number of genes. Among these are α-Sma, Vimentin (Bandopadhyay et al., 2001) and Jag1 (Hofmann and Luisa Iruela-Arispe, 2007). Our own data confirm and extend these reports; in particular, they demonstrate that mural cells at the branching sites express NG2, the most characteristic marker of pericytes to date (Ozerdem et al., 2001), and simultaneously, α-Sma and calponin, which, in mature retinal vasculature, are essentially restricted to VSMCs (Hughes and Chan-Ling, 2004). The basement membrane, which is formed by a synergistic process between endothelial and mural cells, is also modified at Msx1-expressing bifurcation points. A higher density of basement membrane proteins such as versican (which was detected by Ib4) and laminin is observed. Noticeably, Msx1 does not seem to play a role in endothelial secretion of basement membrane proteins since Ib4 and laminin staining in Msx1 ^{flox/ flox} Msx2 ^{flox/ flox} Tie2Cre retinas is completely normal (data not shown). Altogether, our data show that, in addition to mural cells, ECs at branching sites also exhibit specific expression programs as assessed by Msx1 expression.

Hyperaemia (i.e. the increase in blood perfusion associated with neural activity), which is at the basis of functional neuroimaging (Magistretti and Pellerin, 1999), implies mechanisms to regulate blood flow. These have been proposed to take place at the precapillary level in the arteriole (Anderson and McIntosh, 1967). However, the existence of a precapillary sphincter in the retina at the junction between arterioles and capillaries is controversial, has been poorly documented in recent years and the sphincter itself not always rigorously defined (Friedman et al., 1964; Anderson and McIntosh, 1967; Wiedeman et al., 1976; Baez, 1977). From structural analyses, some investigators reported its existence (Benjamin et al., 1998; Ikebe et al., 2001), while other could not detect it (Pannarale et al., 1996). Furthermore, theoretical and experimental studies have shown that regulation of blood flow at the precapillary level is unlikely to play a major role in the capillary filtration coefficient (Bentzer et al., 2001; Boas et al., 2008). Functional analyses indicate that flow regulation takes place over the whole capillary network, via a coupling between astrocytes and blood vessels (Zonta et al., 2003; Cauli et al., 2004; Metea and Newman, 2006; Peppiatt et al., 2006). However, a more recent functional study, based on laser speckle flowmetry, suggests that in the retina, activity-dependent changes in blood flow are controlled largely by arterioles and that capillaries contribute little to them, without documenting specific structures responsible for this control (Srienc et al., 2010). Msx1 expression in the retina is restricted to primary and secondary branching sites between arteries and arterioles. At these sites, contraction-associated proteins are concentrated, and it is plausible that Msx1 may play a role at the arteriolar level in controlling blood flow.

Arteriolar annuli, and also Msx1 expression, could be linked to mechanical constraints at branching sites. Vascular bifurcations are associated with local modifications in rate and pattern of blood flow, including low wall and high oscillatory shear stress (Zarins et al., 1983). Noticeably, the ECs are primarily affected by shear stress, as they are in direct contact with blood flow. Indeed, genes in the endothelium are activated in response to shear stress (Andersson et al., 2005; Ni et al., 2010) via specific DNA responsive elements (Boon and Horrevoets, 2009). In addition, increased hemodynamic stresses in vivo enhance smooth muscle cell coverage of microvessels (Van Gieson et al., 2003). Shear stress is not negligible in retina arteries and arterioles (Ganesan et al., 2010), thus Msx1 expression may be linked to blood flow at branching sites. Noticeably, restriction of Msx1 expression at these sites correlates with the onset of blood flow in the deeper vascular beds of the retina (Stahl et al., 2010), which might substantially change shear stress at bifurcations. It would be necessary to perform arterial obstructive lesions in the retina to evaluate the relation between blood flow and Msx1 expression.

Another intriguing result that we have obtained concerns the Msx1^lacZ pattern of expression in the choroid, which shifts from endothelial to mural cells in the first three weeks of life (Fig. 7). Previously, we published that Msx1 is expressed in adult mouse VSMCs and pericytes and, in the embryo, in ECs and VSMCs (Goupille et al., 2008). However, we have never observed expression of Msx1 in endothelial and mural cell populations in the same vascular bed, either simultaneously or consecutively. Both cells types are regulated by distinct mechanisms and Msx1 has been associated with a number of different cells types in distinct developmental contexts. Therefore, we think that Msx1 plays different and independent roles in the endothelial and mural lineages. In addition, we should stress that, after P21, the Msx1 pattern of expression observed in the choroid is quite similar to the pattern observed in other peripheral arterioles and capillaries (Goupille et al., 2008).

Generation of Msx1^lacZ, Msx2^lacZ and Msx2^flox mutant mice has been described previously (Houzelstein et al., 1997; Lallemand et al., 2005; Bensoussan et al., 2008). The Msx1^flox conditional mutant (Fu et al., 2007) was a generous gift from Dr. Robert Maxson (Los Angeles, California, USA), the Tie2-Cre transgenic mouse (Kisanuki et al., 2001), from Dr. Masashi Yanagisawa (Dallas, Texas, USA). The α-Sm22Cre (Holtwick et al., 2002) and Rosa^mT/mG reporter mouse (Muzumdar et al., 2007) were purchased from The Jackson Laboratory (Bar Harbor, ME, USA). All strains were maintained on an NMRI outbred background. Genotyping primers were previously described (Lopes et al., 2011). Phenotypic analyses were conducted with mutant embryos using littermates as controls. Animals were housed in the Institut Pasteur animal facilities accredited by the French Ministry of Agriculture to perform experiments on live mice (accreditation # B 75 15-05, issued on May 22, 2008), in appliance of the French and European regulations on care and protection of the Laboratory Animals (EC Directive 86/609, French Law 2001-486 issued on June 6, 2001). Protocols were approved by the veterinary staff of the Institut Pasteur animal facility and were performed in compliance with the NIH Animal Welfare Insurance #A5476-01 issued on 02/07/2007.

Eyes were collected at postnatal days 0 to 480 (P0-P480). Eyes were enucleated and immediately fixed with 4% paraformaldehyde in phosphate-buffered saline (PBS) pH 7.2 (Sigma) for 5 to 30 min depending on mouse age. The cornea, lens, sclera, and vitreous were excised by limbal incision under a dissecting microscope. For flat mount preparation, retinas and choroids were detached and separated from the optic nerve with fine forceps. Radial incisions were made towards their edges. The flattened retinas were then washed with PBS and processed for β-gal activity assay and immunohistochemistry. For transverse section analyses, the retinas and choroids were not separated, instead they were immersed in 15% sucrose and OCT compound (TissUE-Tek) before being frozen in liquid N2 and cryostat-sectioned at 20 µm. Immunohistochemistry was performed as described (Lopes et al., 2011).

Msx1^lacZ and Msx2^lacZ genes expression was visualized by Xgal staining as described by (Houzelstein et al., 1997). After staining, tissues were post-fixed in 4% PFA for 1 h at room temperature, then washed 3 times in PBS before immunohistochemistry.

Retinas were permeabilised with PBS containing 0.2% Triton X-100 and 50 mM NH4Cl for 30 min. Tissues were washed 3 times in PBS then treated for 1 h with blocking buffer (1 mM MgCl2, 1 mM CaCl₂, 10% goat serum and 0.5% Tween-20 in PBS). The same blocking solution was used for incubation with primary antibodies overnight at 4°C under gentle agitation. Primary antibodies are listed in Table 1. The retinas were washed four times for 5 min with 0.1% Tween-20 in PBS, incubated for 1 h with secondary antibodies diluted in blocking buffer. Secondary antibodies (Invitrogen) were Alexa Fluor 488 goat anti-mouse and goat anti-rabbit, Alexa Fluor 568 goat anti-rabbit, Alexa Fluor 647 goat anti-mouse, Alexa Fluor 635 goat anti-rabbit at 1/300 and Alexa Fluor 488 streptavidin at 1/1000. Retinas were washed again three times for 5 min with 0.1% Tween-20 in PBS, followed by incubation with 2.5 µg/ml of Hoechst 33342 (bisBenzimide trihydrochloride, Sigma) in PBS for 15 min to label nuclei. They were finally washed in PBS and two times in water and flat mounted with Dako mounting medium.

Whole-mount retinas were primarily observed under a Zeiss Axiophot fluorescence microscope, and a Zeiss Axioplan equipped with an Apotome, and analysed with the Axiovison software (Carl Zeiss, Jena, Germany). Co-localizations were performed with a confocal microscope Zeiss LSM 700 equipped with the Zen software (Carl Zeiss, Jena, Germany). All captured images were assembled using Adobe Photoshop or Adobe Illustrator (Adobe Systems, San Jose, CA, USA). For quantitative analyses, one-way ANOVA was used to compare independent experiments. Comparison between data groups was performed with the non-parametric Dunnett test.

We are very grateful to Dr. Colin Crist for critical reading of the manuscript, to Dr. Robert Maxson and Dr. Masashi Yanagisawa for generously sharing mouse strains. This work was supported by the Institut Pasteur, the CNRS and grants from the French Association pour la Recherche sur le Cancer (ARC) and Ligue contre le Cancer (LCC). Miguel Lopes was the recipient of a fellowship from the Portuguese Fundação Ciência e Tecnologia (FCT).