Sensory evolution in blind cavefish is driven by early embryonic events during gastrulation and neurulation

Animals rely on their sensory systems to perceive relevant stimuli in their environment. Natural selection favors sensory systems that are adapted to stimuli used for survival and reproduction. In line with this idea, there are numerous examples in the literature of sensory specialization in animals according to their habitat. This concerns both external sensory organs and brain areas that process the sensory information. However, not much is known about how existing sensory systems evolve, particularly in terms of size, to fit a specific environment.

To cite a few examples, diurnal rodents have a larger proportion of cerebral cortex devoted to visual areas than nocturnal rodents, the latter having a larger part of their cortex devoted to somatosensory and auditory areas (Campi and Krubitzer, 2010; Krubitzer et al., 2011). Among cichlid fish in African lakes, the relative size of brain regions in different species varies according to the environment and feeding habits (Pollen et al., 2007). Sensory organs also vary greatly among species. Classical cases include the small eyes but highly developed olfactory epithelium of sharks (Collin, 2012), the vibrissae-like tactile hair covering the otherwise hairless skin of the underground-living naked mole rat (Crish et al., 2003; Park et al., 2003; Sarko et al., 2011) or the variation in the number of lateral line neuromasts in sticklebacks depending on their stream versus marine, or benthic versus limnetic, habitat (Wark and Peichel, 2010).

Of note, the evolution of a particular sensory organ is often discussed with no consideration of the role played by other senses, but recent analyses have revealed co-operations and trade-offs among senses (Nummela et al., 2013). Developmentally, this implies that the control of the size of sensory organs is tightly regulated during embryogenesis and later, and that this regulation is somehow coordinated between the different organs.

In cave animals from all phyla, a striking blind (and de-pigmented) phenotype is repeatedly observed. In the blind cavefish (CF) of the species Astyanax mexicanus, which can be advantageously used in developmental comparative analyses because sighted surface fish (SF) of the same species are available (Jeffery, 2008, 2009), the eyes are regressed in adults, but sensory compensations have been reported: CF have more taste buds (chemosensory) (Schemmel, 1967; Varatharasan et al., 2009; Yamamoto et al., 2009) and more head neuromasts (mechanosensory) (Jeffery et al., 2000; Yoshizawa and Jeffery, 2011) than SF. Also consistent with this idea, CF are better at finding food in the dark (Espinasa et al., 2014; Hüppop, 1987) and seem more sensitive to food-related cues than SF (Bibliowicz et al., 2013; Protas et al., 2008). Although the olfactory system (chemosensory) was initially thought to be anatomically and physiologically unchanged in CF (Riedel and Krug, 1997; Schemmel, 1967), later studies found that naris opening is larger in troglomorphic animals (Bibliowicz et al., 2013; Yamamoto et al., 2003), suggesting a possible link between increased food finding capabilities and olfactory specialization. From an evolutionary perspective, these changes in different sensory modules could be either uncoupled or be the result of common selection pressures, acting at developmental and genetic levels (Franz-Odendaal and Hall, 2006; Wilkens, 2010).

Importantly, in Astyanax CF the eyes first develop almost normally during embryogenesis before they then degenerate. The triggering event for eye degeneration is lens apoptosis (Alunni et al., 2007; Yamamoto and Jeffery, 2000). The lack of expression of αA-crystallin, a lens differentiation gene (Behrens et al., 1998; Strickler et al., 2007), probably contributes to the apoptotic phenotype in CF (Hinaux et al., 2014; Ma et al., 2014). However, CF eyes are also smaller than SF eyes from embryonic stages onward (Hinaux et al., 2011; Strickler et al., 2001). What are the developmental mechanisms underlying the regulation of sensory organ size in CF? Previous work has shown that increased Shh signaling and heterochronic Fgf8 signaling in CF have pleiotropic effects on neural development: Shh is indirectly responsible for lens apoptosis (Yamamoto et al., 2004) and also impacts the number of taste buds (Yamamoto et al., 2009), while Shh and Fgf8 influence the relative sizes of the domains of the neural plate and neural tube that are fated to contribute to the retina or to other brain parts (Menuet et al., 2007; Pottin et al., 2011).

Shh and Fgf signaling are also known to affect the development of the placodal region in other vertebrate model species (Bailey et al., 2006; Dutta et al., 2005). The placodes correspond to a region surrounding the neural plate that generates the sensory organs of the head in vertebrates, including the lens, the olfactory epithelium, the otic vesicle and the lateral line (Schlosser, 2006; Streit, 2007, 2008; Torres-Paz and Whitlock, 2014; Whitlock and Westerfield, 2000). Here, we investigated the development of the Astyanax CF placodal region, with a particular focus on the lens and olfactory placodes. We report significant differences in patterning between CF and SF embryos and show that Shh, Fgf and Bmp4 modifications contribute to opposite changes in the size of the lens and olfactory placodes in CF. Further comparing olfactory behavior of the two morphs, we uncover outstanding olfactory skills in CF, confirming their functional olfactory specialization.

We compared placodal field patterning in Pachón CF and SF embryos at the end of gastrulation/beginning of neurulation (Fig. 1). At 10 hours post fertilization (hpf), the shape and size of the neural plate border domain labeled by Dlx3b was similar in SF and CF (Fig. 1A,A′). Inside this Dlx3b-positive border, the anterior neural plate markers Zic1 and Pax6, which label the ʻeye field' and presumptive forebrain, are prominently expressed (Fig. 1B-C′). At the placodal level, the anteriormost part of the Pax6-positive presumptive lens placode territory was lacking in CF embryos, resulting in a smaller lens territory (Fig. 1C-C″). The width (mediolateral extension) of the Pax6 lens placode domain was similar in SF and CF, consisting of 2-3 cell diameters (Fig. 1C-D′). Examination of the anteriormost placodal region using the Dlx3b marker on frontal views also revealed robust differences, with its domain being anteriorly expanded in CF embryos (Fig. 1E-E″, and see below). Thus, early anterior placodal patterning is modified in CF embryos, and the lens placode is reduced in size.

Slightly later, at 12 hpf (end of neurulation) and 16 hpf (mid-somitogenesis), Dlx3b expression was progressively reduced to the presumptive olfactory and otic placode and to the adhesive organ or casquette (Fig. 2A-B′, Fig. S1A) (Pottin et al., 2010). The Dlx3b-negative ventrolateral head region corresponding to the forming eye vesicle was reduced in CF (Fig. 2A-A″), whereas the Dlx3b-positive dorsolateral domain corresponding to the olfactory placode was much larger in CF (Fig. 2B-B″). This was confirmed by the olfactory marker Eya2 (Fig. 2C-C″).

From 20 hpf onwards, the lens mass was clearly identified using Pitx3 (Fig. 2D,D′), which is expressed throughout the lens in both SF and CF (Fig. 2F,G), therefore providing a good proxy for lens size. The Pitx3-positive lens area was much smaller in CF than in SF embryos (Fig. 2D″).

In parallel, the size of the olfactory epithelium (OE) derived from the olfactory placode was assessed after hatching by Olfactory marker protein (OMP) expression (Fig. 2E,E′). The area where OMP-positive neurons are scattered in a salt and pepper pattern was much larger in CF than in SF (Fig. 2E″), and this difference was maintained at 64 hpf (Fig. S2A).

We also examined the anteriormost derivatives of the placodal field, fated to become the adenohypophysis/pituitary in vertebrates (Dutta et al., 2005). Unfortunately, none of the three Astyanax Pitx genes that we cloned was expressed in the pituitary (Fig. S3A-C). We therefore isolated Lhx3, a LIM-homeodomain transcription factor considered as an early and specific pituitary marker. Lhx3 expression starts at ∼24 hpf in Astyanax (not shown). At 28 hpf, the size of the Lhx3-positive domain and the number of Lhx3-positive cells were similar in SF and CF (Fig. S1B), suggesting that, anteriorly, only the olfactory and the lens derivatives vary in size between the two morphs. Finally, the posterior otic placode was identical in size in CF and SF according to Dlx3b expression (Fig. S1A), suggesting that only the anterior placodes are modified in CF.

Altogether, these patterning data suggest that the olfactory placode and epithelium are continuously enlarged in CF at the expense of a reduction of the lens placode and mass.

To test directly a genetic link between the opposite variations in olfactory and lens sizes in CF, we established crosses to generate first (F1) and second (F2) generation hybrids. In 36 hpf F1 hybrids resulting from SF×Pachón CF crosses, the size of the lens (Fig. 3A,A′) and the size of the OE (Fig. 3B,B′) were intermediate between those of a SF and a CF larva of the same age. In 36 hpf F2 hybrids resulting from crosses between two F1 parents, the lens and OE volume were measured concomitantly in single individuals (n=33) on confocal images (Fig. 3C). There was no correlation between lens size and OE size (Fig. 3C′). This suggests that the control of OE size and lens size are complex traits, with a multigenic determinism. These results are in line with previous findings that at least six QTL are involved in the control of lens size (Protas et al., 2007).

We next investigated the origins of the modifications of placodal patterning in CF. That differences were observed as early as 10 hpf suggested that they resulted from modifications during gastrulation. Shh hyper-signaling from the notochord and Fgf8 heterochronic (earlier) expression in the anterior neural ridge are known in CF (Pottin et al., 2011; Yamamoto et al., 2004). In addition, the differences that we observed in the anterior placode region (Fig. 1E,E′) suggested that signaling from the prechordal plate (pcp), an endomesodermal structure with important signaling properties for induction and patterning of the forebrain (Kiecker and Niehrs, 2001), might be affected in CF.

At the end of gastrulation (9.5-10 hpf), the pcp abuts the anterior limit of the embryonic axis, being just rostral to the Shh-expressing notochord and just ventral to the Dlx3b- and Fgf8-expressing anterior neural border (Fig. 4A,B). Bmp4 expression in the Astyanax pcp can be subdivided into the polster (anterior, round in shape, Shh-negative) and posterior prechordal plate (ppcp, elongated in shape, Shh-positive) domains. Bmp4 spatiotemporal expression was compared in the two morphs between 9.5 hpf and 11 hpf (Fig. 4A). In SF, Bmp4 was expressed either in polster only (the majority of embryos) or in polster and ppcp (Fig. 4A-C, Fig. S4A). Conversely, in CF, a majority of embryos showed expression in the ppcp only, and confocal examination confirmed the absence of Bmp4 expression in the polster for many of them (Fig. 4B). Importantly, the polster, as a structure, is present in CF and the migration of the pcp appears similar in the two morphs as (1) a few CF embryos did show some Bmp4 staining in the polster (Fig. 4C, Fig. S4A) and (2) the expression of other genes, such as Pitx1 (Fig. S4C) or the recognized pcp marker goosecoid (Gsc), was present in SF and CF with comparable dynamics (Fig. S5). Thus, significant differences in the dynamics of Bmp4 expression inside the pcp exist between CF and SF (see Fig. 4C legend).

Present and previous results prompted us to investigate whether the multiple changes observed in Shh, Fgf8 and Bmp4 signaling systems are responsible for sensory placode and organ size variations in CF. We designed strategies to assess the influences of Shh and Fgf8, which correspond to quantitative differences in space and time between SF and CF, or the influence of Bmp4, which relates to qualitative differences in expression dynamics inside the pcp between the two morphs.

To test the influence of Shh heterotopy and Fgf8 heterochrony, CF embryos were treated with cyclopamine [an antagonist of the Shh receptor Smoothened (Chen et al., 2002)] or with SU5402 [an antagonist of Fgf receptor signaling (Mohammadi et al., 1997)], respectively, between 6 hpf (shield) and 10 hpf (end of gastrulation), and the sizes of their lens and olfactory placodes were measured at later stages. Cyclopamine (100 µM) resulted in a 19% increase in lens size, as measured at 28 hpf using Pitx3 as marker (Fig. 5A-A‴). Although significant, the cyclopamine-induced increase in lens size resulted in a lens that was still smaller than in SF. Inhibition of Fgf signaling with 0.5 µM SU5402 had no significant effect on lens size (Fig. 5A‴). By contrast, cyclopamine and SU5402 both induced a significant reduction in the size of the olfactory placode (−25% and −21%, respectively), as measured at 16 hpf using the Eya2 marker (Fig. 5B-B‴). Notably, the reduction after both treatments in CF resulted in an olfactory placode of identical size to that of SF embryos.

Because the Bmp4 differences in CF and SF relate to subtle expression dynamics and to the position of the signal within the pcp, their impact was not testable through pharmacological manipulation during a specific time window. An alternative experimental design was used: Bmp4 protein injections were performed at 10 hpf, aiming at anteriorizing or posteriorizing Bmp4 signaling in CF and SF embryos, and thereby mimicking one morph's situation in the other (Fig. 5C). Anterior injections of Bmp4 in CF produced an increase in lens size (+15%; Fig. 5D-D″,F) and a reduction in olfactory placode size (−16%; Fig. 5E-E″,F), suggesting opposite effects of polster Bmp4 signaling on the two sensory derivatives. The CF olfactory placode was also reduced after posterior Bmp4 injection (−16%, Fig. 5E″), suggesting that olfactory derivatives are negatively affected by high levels of Bmp4 signaling at the neural plate stage. Moreover, unlike injections in CF, anterior or posterior injection of Bmp4 in SF did not change lens or olfactory sizes (Fig. 5D″,E″), pointing to the specific lack of Bmp4 in the anterior polster part of the pcp as partly responsible for the small size of the lens in CF, or to the possibility that lens size might already be maximal in SF and therefore cannot be further increased, and that olfactory size might be minimal and cannot be further decreased.

Because an enlargement of the OE in CF may be of functional importance for survival in the dark, we examined whether it remained larger than in SF at juvenile stages. We used the G_αolf olfactory marker on 1-month-old larvae, when the OE is not yet folded inside the naris (Hansen et al., 2004; Wekesa and Anholt, 1999) (Fig. 6A,B,D,E). The size of the G_αolf-positive OE and of the naris opening were significantly larger in CF than in SF (Fig. 6G) (OE, 1.36-fold; naris opening, 1.31-fold; values corrected to body length). Cell counts on sections showed an increased number of cells in the CF OE (Fig. 6H). Labeling of the olfactory projection by insertion of a crystal of lipophilic DiI in the olfactory cup revealed a conserved organization of the olfactory projection onto olfactory bulb glomeruli (Fig. 6C,F). The shape of the OE was consistently round in SF and oval in CF (compare Fig. 6A-F). The olfactory nerve was also longer in CF, which is probably due to the difference in size and shape of the jaw and skull (see also Fig. 7F).

To compare the olfactory skills of 1-month-old juvenile CF and SF (5-6 mm in length), we designed an olfaction assay using amino acids as odorant molecules, in the dark with infrared recordings (Fig. 7A,B, Fig. S6A-C, Movie 1, and see the Materials and Methods). Amino acids are potent feeding cues in teleosts (Byrd and Caprio, 1982; Friedrich and Korsching, 1997) and have been used to test olfactory sensitivity and discrimination (Koide et al., 2009; Lindsay and Vogt, 2004; Vitebsky et al., 2005).

Depending on the fish species and the amino acid, millimolar to nanomolar concentrations trigger neuronal activation in the olfactory system (Dolensek and Valentincic, 2010; Evans and Hara, 1985; Friedrich and Korsching, 1997; Korsching et al., 1997; Vitebsky et al., 2005). We used these concentrations as a starting point. Perfusion of a 10⁻³ M to 10⁻⁵ M stock of alanine or serine resulted in a strong attractive response (positive preference index score) in both morphs (Fig. 7C,D).

Interestingly, when the test was performed in the light and the quantification of the response was carried out for longer, we observed that, in contrast to CF, SF were only transiently attracted to the odorant compartment for ∼4 min, and then swam randomly in the box (Fig. 7E). We interpret this observation as vision interfering with olfactory-driven behavior: SF can see that no food is present despite the food-related odor, and do not persist in their behavioral response. This confirms the importance of the visual sensory modality in controlling SF behavior, and the accuracy of the behavioral set-up.

In both bony and cartilaginous fish, olfaction is used to localize food sources, breeding partners and predators, as well as for communication and learning, while gustation is primarily involved in feeding, including oral processing and evaluating food palatability through direct contact (Collin, 2012; Derby and Sorensen, 2008). Chemosensory detection of amino acids depends on both gustatory and olfactory sensory modalities (Hara, 1994). In most fish species, the detection threshold of the taste buds is in the micromolar range (Hara, 1994, 2006, 2015), higher than the concentrations used here. However, as CF possess more taste buds than SF on their lips and face (Schemmel, 1967; Varatharasan et al., 2009; Yamamoto et al., 2009), we needed to ascertain that the observed response to amino acids was truly olfactory mediated. We chemically ablated the surface of the OE by application of a Triton X-100 solution onto the naris (Iqbal and Byrd-Jacobs, 2010), which resulted in the disappearance of G_αolf immunoreactivity (Fig. 7F). Olfactory responses were measured on bilaterally OE-ablated fish using 10⁻⁵ M alanine, which induces a strong attraction in both morphs in normal fish (see Fig. 7C). OE-ablated SF and CF did not respond to the odor at any time point (Fig. 7G). Importantly, this lack of attraction was not due to a lack of exploratory behavior (Fig. S6D). Thus, at the concentrations used, the behavioral assay measures olfactory-mediated, but not gustatory-mediated, responses.

To determine and compare the threshold concentrations of amino acid detection of the two morphs, we repeated the experiments at progressively decreasing concentrations. Alanine or serine at 10⁻⁶ M still resulted in a robust attraction of CF to the amino acid source, whereas SF no longer displayed such attraction (Fig. 8A,B). The result was identical for the two amino acids, and the lack of response in SF was not due to a difference in swimming exploratory behavior (Fig. S7). As expected, SF did not respond to even lower concentrations (10⁻⁷ M alanine, not shown). To determine the CF discrimination threshold, we further decreased the alanine concentration to 10⁻⁷ M (not shown), 10⁻⁹ M, 10⁻¹⁰ M and 10⁻¹¹ M. Remarkably, the CF detection threshold was 10⁻¹⁰ M (Fig. 7C). Thus, CF are able to respond to 10⁵-fold lower concentrations of amino acids than SF.

Using Astyanax CF as ʻnatural mutants' we uncovered early developmental origins of natural variations in sensory systems. We discuss our findings in terms of a specific understanding of the developmental mechanisms underlying the CF phenotype, and in terms of a more general understanding of sensory development and evolution in vertebrates.

Adult cavefish are blind. During embryonic and larval development, the two main components of their eyes are affected: the retina is small and lacks a ventral quadrant (Pottin et al., 2011) and the lens undergoes apoptosis, which triggers degeneration of the entire eye (Yamamoto and Jeffery, 2000). Moreover, the CF lens enters apoptosis even if transplanted into an SF optic cup (Yamamoto and Jeffery, 2000), suggesting that CF lens defects could stem from early embryonic events. Here, we show that from the earliest possible stage of lens tracing, using transient expression of Pax6 at the pan-placodal stage when placodal precursors are still plastic to give rise to several sensory derivatives (Bailey and Streit, 2006; Dutta et al., 2005; Martin and Groves, 2006; Sjödal et al., 2007), the presumptive territory (and hence the number of precursors) of the lens is reduced in CF as compared with SF embryos.

It seems relevant to CF evolution and adaptation that its olfactory placode is enlarged. In the wild, in the Subterráneo cave (and hence a different CF population to the Pachón used here), adult CF have large nostrils and better chemosensory capabilities than non-troglomorphic fish (Bibliowicz et al., 2013). Here, we show that Pachón juveniles also possess large OEs and outstanding olfactory skills, demonstrating a case of parallel developmental and sensory evolution in two independently evolved CF populations (Bradic et al., 2012). Remarkably, only sharks have been reported to present such sensitivity (low response threshold) to amino acids with, for example, 10⁻¹¹ M alanine eliciting electro-olfactogram responses in the hammerhead shark (Tricas et al., 2009). Although we cannot yet conclude that the difference in olfactory skills of CF and SF is entirely attributable to the difference in olfactory organ size, it is tempting to speculate that it at least in part stems from their developmentally controlled olfactory specialization.

In CF, the increase in olfactory placode size parallels the decrease in lens placode size, suggesting a developmental sensory trade-off: (1) in terms of patterning, the presumptive lens placode territory is reduced anteriorly, at a position that corresponds to the presumptive territory of olfactory precursors according to zebrafish fate maps (Dutta et al., 2005; Toro and Varga, 2007); (2) the experimental manipulation of both Shh and Bmp4 signaling results in concomitant and opposite changes in the size of their lens/olfactory placodes. Together with the case of the gustatory/visual trade-off previously described (Yamamoto et al., 2009), these constitute the only examples to date of a direct link between the development of two sensory organs, involving a pleiotropic effect of these signaling systems, and suggesting indirect selection as an evolutionary driving force underlying the loss of eyes in CF (Jeffery, 2010; Retaux and Casane, 2013). In F2 hybrids, however, we did not observe an inverse correlation between lens and OE size, further suggesting that the developmental trade-off has a multigenic determinism.

We also found that Bmp4 expression and signaling in the pcp are modified in CF compared with SF embryos (Fig. 4). Although several lines of evidence (Pitx1 or Gsc expression) suggest that there is not a problem with migration of the CF pcp, we cannot exclude differences in cell movements between the two morphs. The dynamics of Bmp4 expression are altered, whereas the total level of Bmp4 expression at 10 hpf appears unchanged (Gross et al., 2016). Taking SF as ʻwild type' and CF as ʻmutant', an interpretation of the expression patterns would be the following: in SF, Bmp4 is first turned on in the polster before 9.5 hpf and then in the ppcp; in CF, Bmp4 expression would be turned off prematurely in the polster. Of note, the polster gives rise to the hatching gland, which develops and functions properly in CF [e.g. Pitx1 expression in Figs S3 and 4; Agr2 expression (Pottin et al., 2010); hatching time (Hinaux et al., 2011)]. Therefore, the modification in Bmp4 expression dynamics in CF has no deleterious consequences on polster function, but does influence anterior neural development, as demonstrated by the Bmp4 injection experiments. Coupled with the previously described Shh heterotopy in the notochord and ppcp (Pottin et al., 2011; Yamamoto et al., 2004) and to the Fgf8 heterochrony in the anterior neural territory (Pottin et al., 2011), our data on pcp Bmp4 expression dynamics substantiate that three major signaling systems and organizer centers that orchestrate forebrain and head morphogenesis are altered in CF. Interestingly, Bmp4 is found in QTL intervals controlling eye size (Borowsky and Cohen, 2013; Gross et al., 2008; Protas et al., 2007) and craniofacial bone fragmentation (Gross et al., 2016).

The data obtained from studies in model organisms on the roles of signaling systems in sensory placode development are sometimes contradictory (reviewed by Saint-Jeannet and Moody, 2014), perhaps owing to the fact that activation of similar combinations of signaling molecules at different time points can result in strikingly different outcomes (Lleras-Forero and Streit, 2012; Sjödal et al., 2007). Here, we studied three signaling systems in two morphs of a single species, and investigated their effects on the specification of regional placodal identity and sensory fate at the end of gastrulation/neural plate stage (Fig. 5F,G). In agreement with a stimulatory role of anterior neural ridge-derived Fgf8 signaling at neural fold stage in olfactory fate in chick (Bailey et al., 2006) and an inhibitory role of Shh in lens fate in zebrafish (Barth and Wilson, 1995; Dutta et al., 2005; Karlstrom et al., 1999; Kondoh et al., 2000; Varga et al., 2001), we found that Fgf8 heterochrony and Shh hyper-signaling at the anterior midline are responsible for the enlargement of the olfactory placode and the reduction of the lens placode territory in CF. Experiments performed in chick (Bailey et al., 2006) and Fgfr expression in zebrafish (ZFIN, fgfr1a/fgfr1b expression data) support a potential direct effect of Fgf signaling on the placodal region. Our results on CF, in which the difference with SF in terms of Fgf8 signaling comprises a 1.5 h expression heterochrony at the anterior neural border, points to the importance of the timing of Fgf8 signaling on placodal tissue [of note, Fgf3 expression is unchanged in CF (Pottin et al., 2011)]. Conversely, Shh signaling effects on the lens and olfactory placode are probably indirect (Dutta et al., 2005), possibly via Fgf8 in the case of the olfactory placode (Pottin et al., 2011).

We also propose that Bmp4, as a signaling molecule secreted from the pcp as a signaling center, may influence placodal cell fate (Fig. 5F,G). This finding is distinct from the well-established early role of Bmp signaling in the specification of the neural plate border region/preplacodal region, which involves Bmp2/4/7 activity from the ventral side of the embryo at late blastula stages and the subsequent inhibition of Bmp signaling from the epidermis by Bmp inhibitors during gastrulation (Ahrens and Schlosser, 2005; Kwon et al., 2010; Nguyen et al., 1998; Reichert et al., 2013; Saint-Jeannet and Moody, 2014). In chick, a direct role for Bmp signaling in placodal precursor specification at late gastrula stage, as well as a role for the time of Bmp exposure in the decision to follow a lens versus an olfactory fate, have been described (Sjödal et al., 2007). Our comparative approach using CF and SF have revealed a difference in Bmp4 expression dynamics in the pcp. This result and the functional experiments employing local Bmp4 injections at 10 hpf suggest that the timing of Bmp4 signaling from the pcp might be an important cue for lens fate specification, and support a general negative effect of sustained Bmp4 anterior midline signaling on olfactory fate. Our findings are also consistent with Bmp4^−/− mice having normal olfactory derivatives but lacking lenses, although their placodal progenitors are initially correctly specified (Furuta and Hogan, 1998). Finally, the CF Bmp4 phenotype is not fully penetrant (Fig. 4). This might result from a polymorphism that is not fixed in CF. Yet, all CF have smaller lenses, probably because other pathways (Hh, Fgf, and perhaps others yet to be discovered) still contribute to size control, regardless of the Bmp4 phenotype.

Finally, our results concerning the pituitary are surprising. As Shh signaling is a potent inducer of pituitary fate (Dutta et al., 2005; Herzog et al., 2004; Karlstrom et al., 1999; Kondoh et al., 2000; Treier et al., 2001; Varga et al., 2001), one could expect to find a large adenohypophysis in CF with Shh hyper-signaling (Yamamoto et al., 2004). This was not the case. Indirectly, this suggests that other signaling modifications might compensate for Shh hyper-signaling in CF, and points to a possible negative control of pituitary fate in the placode by Fgf8 from the anterior neural ridge and Bmp4 from the pcp.

None of our experimental treatments led to full ʻrecovery' of CF lens to a size comparable to that of the SF lens. This is probably due to the fact that both Shh hyper-signaling from the ventral midline and lack of Bmp4 signal from the pcp are responsible for the small lens size in CF (Fig. 5F,G). Conversely, inhibition of either Shh or Fgf signaling in CF resulted in an olfactory placode identical in size to that of SF embryos. This shows the importance of Fgf8 and the associated Fgf8/Shh autoregulatory loop in the control of olfactory fate. These observations illuminate the subtle equilibrium that must exist in space and time between the signaling systems to orchestrate the development of the surrounding sensory epithelium. In model species, manipulation of early signaling systems usually results in ʻmonstrous' phenotypes. It might thus seem doubtful that morphological evolution is due to modifications at this level. However, we show that subtle changes in the equilibrium between signaling systems at the end of gastrulation can participate in natural morphological evolution, and are part of the developmental evolutionary toolkit. Here, we have deciphered the impact of fine changes in the strength or timing of three signaling pathways emanating from three organizer centers in the developmental evolution of sensory systems in Astyanax. As we have studied the CF ʻnatural mutant', an animal that is viable and adapted to its environment, the early developmental mechanisms that we have uncovered are probably generally applicable and relevant to adaptive sensory evolution and specialization in vertebrates.

Laboratory stocks of A. mexicanus SF and Pachón CF were obtained in 2004 from the Jeffery laboratory at the University of Maryland and maintained as previously described (Elipot et al., 2014). Embryos were collected after natural spawning, grown at 23°C, staged according to the developmental staging table (Hinaux et al., 2011) and fixed in 4% paraformaldehyde. After progressive dehydration in methanol, they were stored at –20°C.

Animals were treated according to French and European regulations for handling of animals in research. S.R.'s authorization for use of animals in research is 91-116, and Paris Centre-Sud Ethic Committee authorization numbers are 2012-0052 and 2012-0055.

F1 hybrid larvae were obtained by in vitro fertilization of female SF eggs by Pachón male sperm (Elipot et al., 2014). They were photographed at 36 hpf under an Olympus SZX16 stereomicroscope. Measurements were made on the images using ImageJ software (NIH).

F2 larvae were fixed and immediately double stained with DAPI and DiI and imaged under an SP8 confocal microscope (Leica). Lens and OE volumes were measured with Fiji using the MeasureStack plugin.

Total RNA from Astyanax embryos of various stages (6-36 hpf) was reverse-transcribed using the iScript cDNA synthesis kit (Bio-Rad). Partial cDNA sequences for Omp (GenBank ID KP826791.1), Lhx3 (KP826792.1), Pitx1 (KP826793.1) and Pitx2 (KP826794.1) were amplified by PCR (for primers, see the supplementary Materials and Methods) and subcloned in the TOPO-PCR II vector (Invitrogen). Zic1 (FO290256) and Eya2 (FO211529) partial cDNAs originate from our cDNA library (Hinaux et al., 2013). Phylogenetic analyses (Figs S8-S12) were conducted to determine orthology relationships. Deuterostomes sequences were retrieved from the Ensembl database based on their annotation, aligned using MAFFT v7.023b (http://mafft.cbrc.jp/alignment/software/) with manual correction. Maximum likelihood analyses were performed using PhyML (Guindon and Gascuel, 2003), with the LG model of amino acid substitution and a BioNJ tree as the input tree. A gamma distribution with four discrete categories was used. The gamma shape parameter and the proportion of invariant sites were optimized during the searches. The statistical significance of the nodes was assessed by bootstrapping (100 replicates). Shh (AY661431), Pax6 (AY651762.1) and Bmp4 (DQ915173) cDNA were isolated previously.

Digoxygenin-labeled riboprobes were synthesized from PCR templates. A protocol for automated whole-mount in situ hybridization (Intavis) was used (Deyts et al., 2005). Embryos were photographed in toto under a Nikon AZ100 stereomicroscope using agarose wells. Some were embedded in paraffin, sectioned (8 µm) and counterstained, using Prolong Gold anti-fade with DAPI (Invitrogen).

For fluorescent double in situ hybridization, Cy3- and FITC-tyramides were prepared and embryos were processed as previously described (Zhou and Vize, 2004; Pottin et al., 2011). Embryos were imaged with either an Olympus SZX16 stereomicroscope, a Zeiss Apotome or a Nikon Eclipse E800 microscope.

Whole-mount immunofluorescence was performed using G_αolf primary antibody (Santa Cruz Biotechnology, sc-383; 1/1000) and Alexa Fluor secondary antibodies (Invitrogen, A-11008 and A-11037; 1/500). Samples were imaged using an Olympus SZX16 microscope. Size measurements were performed on images using ImageJ.

Manually dechorionated CF embryos were incubated in 100 µM cyclopamine (C-8700, LC Laboratories) or 0.5 µM SU5402 (215543-92-3, Calbiochem) diluted in blue water (Elipot et al., 2014) from 6 to 10 hpf. Controls were incubated in an equivalent concentration of ethanol or DMSO, respectively. They were washed in blue water and fixed at 16 hpf or 28 hpf. To define and ascertain the working concentrations of cyclopamine and SU5402, we checked that hatched larvae have a typical ʻcomma shape' or tail bud defects, respectively.

CF and SF embryos at 10 hpf were placed in agarose wells under an Olympus SZX12 stereomicroscope equipped with a micromanipulator. They were micro-injected with a solution of 0.1 µg/µl Bmp4 protein (R&D Systems), 10% glycerol in Phenol Red/water, either anteriorly to the polster region or under the notochord, posterior to the head. Embryos were photographed under an Olympus SZX16 stereomicroscope and sorted according to the precise region of injection (marked by Phenol Red). They were fixed at 16 hpf or 28 hpf.

Four juveniles were placed in each of two behavioral testing boxes positioned on an infrared light source (ViewPoint). They were acclimatized for 2 h in the dark. Perfusion of the amino acid and control solution was initiated simultaneously as recordings started (Dragonfly2 camera, ViewPoint imaging software). A preference index score was calculated for each fish every 30 s, depending on its position relative to the amino acid source: +3 for the quadrant closest to the source, with +1, −1, −3 for the quadrants progressively further from the source (Fig. 7A; supplementary Materials and Methods). This score was then corrected for the position of the fish when the amino acid reached the box at 1.5 min after perfusion opening.

Ablation of the OE was adapted from Iqbal and Byrd-Jacobs (2010). One-month-old fish were anesthetized using 0.1% MS-222 (Sigma) in embryo medium (EM) and chemical ablation of each naris was performed using a solution of 0.05% Methylene Blue, 0.7% Triton X-100 in EM perfused continuously with a micro-injector (Eppendorf, Femtojet) for 90 s. After ablation, fish were allowed to recover for 24 h before being used for behavioral tests or fixed for immunohistochemistry.

We thank Stéphane Père and Diane Denis for Astyanax care; Laurent Legendre and Victor Simon for obtaining hybrids.