Molecular regionalization of the developing amphioxus neural tube challenges major partitions of the vertebrate brain

Molecular markers define and regionalize the amphioxus floor plate

AP and DV subdivisions in developing chordate neural tubes are defined according to axial references. Conventionally, such references are provided in vertebrates by the axial mesoderm (the notochord), the floor plate, roof plate, and alar–basal boundary within the lateral walls of the neural tube, all of which are topologically parallel to each other (Fig 1). Amphioxus has a notochord, which extends singularly beyond the forebrain [65], and a floor plate [66–70]. As previously reported for the Floridan amphioxus Branchiostoma floridae [71,72], we observed in the European amphioxus B. lanceolatum that the gene FoxA2-1 is a selective marker of the notochord (Fig 2A–2A′′′, 2E and 2F), while Nkx2.1 seems to be a general floor plate marker at the 7-somite neurula stage (Fig 2B–2B′′′). As in vertebrates, in which its expression in the floor plate is transient [73], Nkx2.1 expression is highly dynamic during amphioxus CNS development (S1 Fig). Nkx2.1 is observed along the entire presumptive floor plate at early- and mid-larval stages, but it subsequently becomes restricted rostralwards.

Since the whole neural tube of amphioxus sits on top of the notochord, it should be, in theory, regarded as topographically epichordal. Thus, as the floor plate is induced vertically by the notochord [74–76], we a priori expected the amphioxus floor plate to extend all along the acroterminal neural midline (up to the neuropore), in contrast to the vertebrate floor plate, which stops at the mamillary pouch of the hypothalamus, coinciding with the approximate position of the rostral tip of the notochord (Fig 1A and 1B; [2,4]). Instead, we observed that the floor plate, defined by Nkx2.1 expression, does not reach the anterior neural border, but it ends in a slightly expanded median patch that recalls the mamillary hypothalamic ending observed in vertebrates (Fig 2B and 2B′ insets, K; see also [66]). Interestingly, Hedgehog (Hh), which is a well-established floor plate marker in vertebrates [74,77,78], and Nkx6 are also expressed in the amphioxus floor plate, but their anterior limit of expression is more caudal than that of Nkx2.1 (Figs 2C–2C′′′, 2K, 2L and 7D–7D′′; a similar expression for Hh has been reported in B. floridae [69]). Goosecoid (Gsc) is also expressed in the floor plate (in contrast to previous reports [79]) in a variable and patchy pattern that might reflect cyclic dynamic changes (Fig 2D–2D′′′). These markers differentiate two major floor plate AP regions: (i) a rostral-most median floor domain characterized by only Nkx2.1 expression, which corresponds to the floor plate of the forebrain region that we refer to as the amphioxus hypothalamo-prethalamic primordium (HyPTh; see below and Fig 2K and 2L); and (ii) the rest of the floor plate, defined by Hh, Gsc, Nkx6, and Nkx2.1 expression.

Molecular heterogeneity of the amphioxus axial mesoderm: Notochord and a possible prechordal primordium

We next examined the genoarchitecture of the amphioxus axial mesoderm to assess the existence of a putative prechordal plate homolog. According to the updated prosomeric model [2,4], the latter tissue lies topologically rostral to the neural primordium and the notochord (Fig 1B). As mentioned above, FoxA2-1 labels the whole amphioxus prospective notochord (Fig 2A–2A′′′). On the other hand, the expression of both Hh and IrxB in the axial notochordal tissue does not reach the rostral tip of the FoxA2-1–positive domain, stopping beneath the rostral end of the Nkx2.1–positive HyPTh floor plate (Fig 2L, 2E–2J; it should be noted, however, that IrxB expression seems to reach the anterior tip of the notochord in B. floridae [80]). Moreover, in the amphioxus axial mesoderm, Six3/6 expression was observed exclusively in the rostral tip of the FoxA2-1–positive domain, beyond the Hh/IrxB–positive part of the notochord (Fig 2K, 2L and 2J; see also [81]). Interestingly, this Six3/6 expression is maintained at later stages, when the notochord is fully formed [81], indicating that its rostral tip has a distinct molecular signature compared to the rest of the notochord. Remarkably, in vertebrates, Six3 is expressed in the prechordal plate but not in the notochord at any level [24]; therefore, the rostral notochordal tip of amphioxus might represent a possible prechordal plate homologue, previously unrecognized due to its histologic similarity to the notochord proper (see Discussion).

Finally, we found that Gbx expression appears restricted to a more caudal sector of the notochord, whose rostral border is posterior to the caudal boundary of the HyPTh neural domain (Figs 2K, 2L, 3F and 3L inset). Previous studies [37] and other observations described below suggest that the Gbx-expressing domain of the notochord and overlying neural tissue begins at the rostral end of the major region we term Rhombencephalo-Spinal primordium (RhSp; Fig 2L). In summary, we observed that the amphioxus axial mesoderm is subdivided molecularly into various regions, which have direct correspondence with major subdivisions in the overlaying neural plate.

The incipient neural tube of amphioxus possesses distinct floor, basal, and alar plates

Previous gene expression studies provided evidence for the presence of longitudinal zones positioned parallel to the floor plate, implying DV patterning in the amphioxus CNS [69,82]. We thus investigated the extent of DV regionalization and its related boundaries by systematically searching for gene expression patterns with specific DV domains. We found that most of the examined patterns could be classified into three groups (Figs 3–9): (i) peripheral genes, with expression restricted to the periphery of the neural plate (future topologically dorsal or alar zone; Six3/6, Lhx2/9b, Zic, Msx, Pax2/5/8, Pax3/7, Nova); (ii) internal genes, with expression domains respecting the former peripheral longitudinal zone (Pou3f, Sim, FoxD, Meis, Lef, Lhx1/5, Hox3, Hox6, FoxB); and (iii) pan-DV genes, expressed across both aforementioned domains (Otx, Gbx, Fezf, Irx, Pax4/6, Six3/6, Nkx2.2, Meis, Rx, Hox1, Wnt3, Wnt7, Nova, Ebf). A few markers were ascribed to two of these categories since they have DV expression subdomains that differ depending on the AP partition in which they are expressed (see below and Fig 9). Altogether, these patterns suggest the existence of continuous basal and alar plate zones that extend longitudinally throughout the amphioxus neural tube primordium. As in vertebrates, the right and left moieties meet frontally around the rostral end of the floor plate (Fig 10A), as clearly exemplified by the alar expression of Lhx2/9b (Fig 6C′ and 6C′′).

Conserved Otx and Gbx expression patterns define a primary AP partition

In all vertebrates, dynamic antagonistic expression of Otx (rostral) and Gbx (caudal) in the neural plate eventually reaches an equilibrium at the caudal end of the midbrain, defining the MHB (Fig 3A and 3D; [5,83–88]). Clonal labeling studies performed in frogs at the 64 blastomere stage showed that this is the earliest detectable brain transverse boundary [89]. A comparable boundary is also present in amphioxus, aligned between the first and second somites [37], which we further corroborated at the 7-somite neurula stage in B. lanceolatum (Fig 3B–3F). Accordingly, it was suggested that the first intersomitic limit of amphioxus roughly marks the genetic homolog of the MHB of vertebrates [37,52,53,58,90]. Thus, it can be postulated that these early expression domains in both lineages define the boundary between a rostral Otx-positive “archencephalic prototagma” (ARCH; Fig 3J and 3M) and a caudal Gbx-positive “deuteroencephalic prototagma” (DEU; Fig 3J and 3M). In addition to Gbx, several other amphioxus genes show specific expression within DEU at this stage, abutting rostrally the ARCH–DEU boundary (Wnt3, Wnt7, FoxB, Pax2/5/8, and Msx; Figs 6G–6H′′ and 7E–7G′′).

The archencephalon is regionalized anteroposteriorly into the HyPTh and DiMes regions

We previously showed that the ARCH domain can be subdivided anteroposteriorly based on Fezf and Irx expression [36]. In both vertebrates and amphioxus, Fezf genes are expressed in the rostral-most part of the CNS at early neural tube stages, creating an anterior subdomain within the Otx-positive territory (Fig 3G–3I) and thus leaving a gap between the caudal end of their expression and the start of that of Gbx in DEU (Fig 3K and 3L). On the other hand, Irx genes are expressed within this gap, abutting rostrally with Fezf and extending posteriorly into the Gbx-positive DEU tagma (Fig 3K–3O). Studies in Xenopus, zebrafish, and mice, comparing Fezf and Irx expression patterns with fate mapping data, have shown that the transverse Fezf-Irx interface marks the prethalamo–thalamic boundary where the ZLI will develop [21,91–94]. Based on the expression patterns observed in the 7-somite amphioxus neurula, we accordingly defined a rostral Fezf-positive HyPTh and a caudal Irx-positive DiMes intercalated between the Fezf-positive and Gbx-positive domains (Fig 3K–3O). Remarkably, several genes, including Pax4/6, Six3/6, Pou3f, and Sim, are expressed specifically or most strongly within the DiMes (Fig 4, see also Fig 8A–8C′′), supporting the distinct identity of this region. Moreover, other genes in addition to Fezf appear restricted to HyPTh (e.g., Rx throughout it, Fig 5B and 5B′′, and FoxD in its basal plate subdomain, Fig 5C and 5C′′; see also Figs 8D–8E′′ and 9) or have distinct expression subdomains within HyPTh (e.g., Nova, S2C and S2′′ Fig). On the other hand, other markers, such as Ebf, are expressed caudally to the Fezf/Irx limit, similarly to the three Irx genes (IrxA-C)(Figs 3N, 3O, 4G, 4G′, 6A and 6B′′).

Triple fluorescent in situ hybridization and confocal 3-D reconstruction show that the Fezf-positive HyPTh, the Pax4/6-positive DiMes and the Gbx-positive RhSp domains abut sharply one another. Interestingly, the intermediate domain, DiMes, is very small, consisting only of two rows of cells along the AP dimension (Fig 8A–8B′′). Analogous fluorescent in situ hybridization comparison of Fezf, Six3/6, and Gbx patterns shows that Six3/6 is also strongly expressed in the DiMes compartment (Fig 8C–8C′′). The Fezf and Gbx markers are expressed with similar mutual relationships also at the 4/5-somite (early neurula) stage (S3A–S3B′ Fig), leaving an expression gap where weak Six3/6 signal can already be detected (S3E–S3F′ Fig). Therefore, both the ARCH/DEU limit and the HyPTh and DiMes subdivisions within ARCH are established very early in amphioxus CNS development.

The HyPTh is divided into three AP domains

We next sought to identify further AP molecular partitions within the HyPTh and DiMes forebrain domains of the 7-somite neurula. Unlike the DiMes, for which we could not identify any molecular subdivision, eight examined markers showed restricted expression domains within HyPTh, sometimes limited to either alar or basal regions. Their combined pattern is consistent with the existence of three AP subdivisions within the HyPTh, which we termed Rostral HyPTh, Intermediate HyPTh and Caudal HyPTh (Rostral-HyPTh, Interm-HyPTh, Caudal-HyPTh; Fig 10A; see Discussion for possible homology relationships with partitions in the vertebrate forebrain). For instance, the expression of six rostral markers (Nkx2.2, Nova, Meis, Pou3f, Lef, Lhx2/9b) appears across Rostral-HyPTh and Interm-HyPTh, but seems to respect a transverse double row of cells that lie anterior to the Irx-expressing DiMes; topologically, this caudal negative domain of HyPTh (Caudal-HyPTh) would correspond in vertebrates to the primordium of the prethalamic region (Fig 10A). This partition can be visualized as a gap of negative labeling, e.g., by double in situ hybridization for Meis and IrxC or for Nkx2.2 and IrxB (Figs 5G–5G′ and 8F–8F′′, respectively).

While Nkx2.2 and Nova signals are present at both alar and basal levels of Rostral-HyPTh and Interm-HyPTh (but not in the corresponding part of the floor plate; Fig 5D–5D′′ and S3C–S3C′′ Fig; see schematic details of expression in Fig 9), Pou3f, Meis, and Lef expression is restricted to the local basal region (also respecting the floor plate; Figs 4F–4F′′, 5F–5F′′ and 5H–5H′′, respectively), and Lhx2/9b expression appears selectively in the peripheral alar region (Fig 6C–6C′′). Six3/6 is the only studied marker restricted to the Rostral-HyPTh, specifically in the alar plate (Figs 2J, 4D–4D′′ and 8C–8C′′; similar expression of Six3/6 in the anterior-most part of the neural plate was also reported in B. floridae [81]). On the other hand, Zic, which is a well-known marker of the alar and roof plates in the CNS of vertebrates [95], is expressed throughout the presumptive alar plate of the HyPTh, but its expression is significantly stronger at the Caudal-HyPTh, showing decreasing signal towards the rostral alar parts of the HyPTh complex (S2G–S2G′′ Fig). In contrast to the major HyPTh and DiMes partitions, the three HyPTh molecular subdivisions are not fully established at the 4/5-somite stage (S3 Fig).

Anteroposterior regionalization of the amphioxus DEU domain

Numerous amphioxus genes have been previously reported to show iterative expression domains within the DEU, suggesting the existence of characteristic subdivisions within this partition ([96–99], and see S1 Table). Consistent with these studies, we identified several molecular AP partitions within the rostral-most subdomain of DEU, referred to here as the RhSp, which roughly ends caudal to the fifth somite at the 7-somite neurula stage (Figs 9C and 10A). Gbx, Wnt3, Wnt7, and FoxB appear selectively expressed throughout the RhSp; all of them abut rostrally the DiMes/RhSp boundary, and their expression domains end at different caudal levels, either coinciding with the end of somite five or extending further caudalwards (Figs 3E, 3F, 6G–6G′′, 6H–6H′′ and 7E–7E′′). Gbx, Wnt3, and Wnt7 occupy both alar and basal regions (but not the floor plate), as previously described [37,100,101], while FoxB is restricted to basal areas (Fig 7E; see also [102]). Hox1, Hox3, and Hox6 genes are also expressed along alar and basal parts of the RhSp, with rostral expression borders that correspond with the intersomitic limits S3/S4, S4/S5, and S5/S6, respectively (Fig 7A–7C′′; see also [53–57]). As mentioned above, these molecular partitions are complemented by patterns of iterated spots with negative intervals, which can be aligned with the center (Lhx1/5 and Pou3f, Figs 6E–6E′′ and 4F–4F′′) or posterior half of the somites (Nova, S2C–S2C′′ Fig) or the inter-somitic boundaries (Pax3/7, S2E–S2E′′ Fig). In the case of Pax2/5/8, patches are less well defined, particularly caudally, where they become nearly continuous (Fig 7F–7F′′). Finally, some genes show isolated spots of expression located at different positions within the RhSp AP subdivisions: Msx (Fig 7G–7G′′), Meis (Fig 5F–5F′′), Zic (S2G–S2G′′ Fig), Nkx6 (Fig 7D–7D′′), and Otp (Fig 6F–6F′′), sometimes correlating with the prospective position of the future pigmented photoreceptor spot.

Experimental suppression of the ZLI and IsO organizers in vertebrates alters di-mesencephalic patterning and generates a remnant that resembles the amphioxus DiMes

A major implication of our comparison of the overall CNS genoarchitecture between amphioxus and vertebrates is that the small amphioxus Pax4/6-positive DiMes corresponds topologically to the large vertebrate region comprising the thalamus, pretectum, and midbrain (Fig 10A–10B′′, see Discussion). Patterning of this territory in vertebrates occurs under the dual control of the secondary brain organizers (ZLI and IsO, see Introduction), which induce the molecular subdivision and differential growth of an initially Pax6-positive primordium. Among other effects, these organizers inhibit the expression of Pax6 at the two ends of the territory so that Pax6 signal becomes restricted to the caudal pretectum (p1) and the epithalamus (dorsal-most part of p2)(Fig 11C′). Accordingly, we hypothesized that the small size and the lack of internal regionalization of the DiMes, particularly with respect to Pax4/6 expression, may be, at least in part, related to the absence of ZLI-like and IsO-like effects in amphioxus.

To gather support for this hypothesis, we turned first to loss-of-function transgenic mouse lines in which either the ZLI or the IsO are absent. Double Fezf1^-/-Fezf2^-/- mutants [103] lack the ZLI organizer and largely lose the molecular identity of the alar thalamic field, displaying expanded expression of the pretectal markers Pax6⁺ and Ebf1⁺ (Fig 11B and 11B′–11B′′′′); the midbrain was not altered in these mice. We also studied En1^cre/+; Fgf8^flox/flox mice (see Materials and Methods) in which the IsO is deleted across the MHB [104]. The resulting phenotype showed a reduction of the AP dimension of the pretecto-mesencephalic region down to one-third of its normal size and an abnormal caudal expansion of PAX6 immunoreaction (as well as of the posterior commissure), suggesting a lack of differential specification of the midbrain versus the pretectum (Fig 11C and 11C′–11C′′′′); in this case, the thalamus seemed normal.

Next, we tried to eliminate both the ZLI and the IsO together in zebrafish, using quadruple morpholino (4MO) treatment against otx1a, otx2, eng2a, and eng2b. Although effects in neural progenitors of other areas cannot be ruled out, double morpholinos against otx1a and otx2 were successfully used previously to specifically abolish the ZLI [105], while double morpholino treatment against eng2a and eng2b caused the loss of the IsO [106], and expression of pax6a throughout the midbrain remnant [107]. Strikingly, the normal pax6a-negative gaps corresponding to the alar plate of the midbrain and diencephalic thalamus were abolished or severely reduced in nearly all (84%) 4MO specimens tested (Fig 11F; n = 75, p = 3.84 × 10⁻³¹, one-sided Fisher Exact test), often resulting in a continuous expression of pax6a between the rostral conserved part of the forebrain and the hindbrain (Fig 11G, 11H, 11K and 11L and sagittal sections in insets in Fig 11H and 11L). Supporting the effective suppression of the two organizers in this experiment, we observed disappearance of the dorsal ZLI spike expression of shha and of the MHB-related transverse band of wnt1 expression (Fig 11D, 11E, 11H, 11I, 11L and 11M). In addition, all 4MO embryos showed a significant reduction of the zebrafish DiMes-like remnant at 28 h post fertilization (hpf) compared to the controls (Fig 11G′ and 11K′).

Integrative genoarchitectonic model of the amphioxus incipient neural tube

Consistent with previous results [37,59,64], our data show that the incipient amphioxus neural tube is molecularly divided anteroposteriorly into a rostral archencephalic (ARCH) and a caudal deuterencephalic (DEU) portions from very early stages, similarly to vertebrates (Fig 10A and 10B). Traditionally, three main AP divisions are defined in the vertebrate ARCH (Fig 1): the secondary prosencephalon (encompassing hypothalamus plus telencephalon), the diencephalon, and the midbrain. On the contrary, the ARCH of amphioxus shows only two main divisions, which we termed DiMes and HyPTh (Fig 10B). DiMes is a small caudal region consisting of two rows of cells that occupies the topological position corresponding in vertebrates to the midbrain and the two diencephalic segments that lie caudal to the ZLI organizer; no internal subdivisions were detected within DiMes. In contrast, the HyPTh encompasses three molecularly distinct segments: a relatively large, bipartite, putative hypothalamus-homolog region (where neither telencephalic nor optic vesicles are present [113]) plus a caudal region that occupies the topological position corresponding to the vertebrate prethalamus. In the case of the DEU, its rostral portion, referred to here as RhSp primordium, may represent a field-homolog of the vertebrate hindbrain, and shows a number of gene expression patterns that configure periodic segment-like territories (Fig 9C).

Notably, these major AP partitions of the developing CNS are mirrored by molecularly defined subdivisions in the underlying axial mesoderm. Indeed, we provide evidence that the distinct molecular entity at the rostral tip of the amphioxus notochord may be homologous to the vertebrate prechordal plate, being thus essentially different from the notochord proper that underlies the brain floor plate (Fig 2). This potentially prechordal region lies topologically rostral to the HyPTh (not under it), as occurs with the prechordal plate in vertebrates [2,4], and is characterized by the absence of Hh and Nkx6 and the specific expression of Six3/6, which is also characteristic of the vertebrate prechordal plate [24]. Therefore, it is possible that this special notochord-looking region—which also shows unusual proliferation and rostralward growth [70,114]—may correspond to a variant prechordal plate homolog and/or plays partly equivalent signaling functions to this structure in amphioxus, despite the absence of some key vertebrate prechordal markers (Gsc, noggin, and chordin [79,115]).

Finally, regarding DV patterning, multiple markers provide extensive evidence for continuous molecularly distinct floor, basal, and alar zones throughout the length of the incipient neural tube (Figs 9 and 10). Although we did not find selective markers for the roof plate, it is possible that these may exist at later stages, upon neural tube closure. Consistent with the idea that the alar plate and roof plate are not differentially specified in amphioxus at these stages, orthologs of several vertebrate neural plate border makers (e.g., Pax3/7, Msx, and Zic) were found to be expressed broadly in the alar plate (Fig 9; observed also in B. floridae [116]).

Possible homology relationships between HyPTh partitions and vertebrate forebrain neuromeres

According to the updated prosomeric model [4], the nontelencephalic part of the vertebrate secondary prosencephalon can be subdivided into two main neuromeres: terminal (THy, hp2) and peduncular (PHy, hp1) hypothalamic prosomeres. In addition, THy includes a specialized rostral-most median part extending dorsoventrally, the acroterminal area [2,4,117](Fig 10C′′). This molecularly distinct domain produces a number of specialized formations along the DV axis, including the alar preoptic lamina terminalis, the optic chiasma, the eye vesicles, the basal median eminence, and the neurohypophysis.

In amphioxus, HyPTh represents a relatively large, molecularly distinct forebrain region lying rostral to the DiMes. This domain has specific expression of Fezf throughout (Fig 9), which is also absent caudal to the ZLI limit in vertebrates [93,103,118]. Our analysis suggests that there are three molecularly distinct AP subdivisions within the amphioxus HyPTh, which we termed Rostral-HyPTh, Interm-HyPTh, and Caudal-HyPTh. By direct topological ascription, these might correspond, respectively, to the transverse THy (including a rostromedian acroterminal region) and PHy hypothalamic segments and a prethalamus-like segment next to the DiMes.

Six3/6 was the only studied marker that selectively labeled Rostral-HyPTh. Remarkably, in mice, Six3 is expressed extensively dorsoventrally across the alar and basal zones of THy (including the acroterminal area), whereas Six6 signal is restricted to a ventral suprachiasmatic part of the THy acroterminal alar plate, but none of them are expressed at PHy [2,117]. These data support a genetic equivalence between the Rostral-HyPTh and THy, in addition to their topological correspondence. Moreover, amphioxus develops in its acroterminal region (orange domain in Fig 10C and 10C′) a median primordial eye patch and, ventral to it, a median group of “infundibular cells” [90], which are located above the most anterior floor plate cells (gray cells in Fig 10C) and might represent a homologue of the vertebrate neurohypophysis. As mentioned above, in vertebrates, both the eyes and the neurohypophysis develop from the acroterminal area [2], further supporting the homology of vertebrate and amphioxus acroterminal domains and thus of Rostral-HyPTh and THy (Fig 10C–10C′′).

In the case of the Caudal-HyPTh primordium, its topological position, lying directly rostral to the Fezf-Irx boundary, provides grounds to suggest field homology with the vertebrate prethalamus. Importantly, previous studies indicate that Fezf genes are essential to specify the prethalamic domain in vertebrates; however, unlike regions within the vertebrate DiMes counterpart (see below), this specification is independent of the ZLI organizing activity and occurs prior to its formation [93,103] and is thus compatible with the amphioxus scenario at the examined stage. Nonetheless, it should be noted that, although more weakly expressed, the presence of Rx expression in Caudal-HyPTh (absent in the prethalamus of vertebrates [118]), suggests the alternative possibility that this partition may represent a primordium homolog to both the peduncular hypothalamus and prethalamic region.

Close developmental and evolutionary relationship of thalamus, pretectum, and midbrain

One of the most striking implications of our results is that the small, Pax4/6-positive DiMes of amphioxus corresponds topologically to the region comprising the vertebrate thalamus, pretectum, and midbrain (Fig 10). While this area is not subdivided in amphioxus and consists only of two cell rows at the neurula stage, the equivalent vertebrate region shows three major partitions and extensive cell proliferation. These partitions in vertebrates originate during development as a consequence of the action of the secondary brain organizers on a Pax6-positive primordium. In particular, Shh signaling from the ZLI is crucial for the specification of the thalamus [6–8,119], and Fgf8 and Wnt1 expression from the IsO are necessary for proper midbrain specification and differential caudal growth [5,10,12,13,120–123]. Moreover, due to the action of these organizers, the expression of Pax6 in this primordium is mainly restricted to the pretectum and the epithalamus and becomes absent in the ventricular zone of the thalamus and midbrain (Fig 11) [124,125].

Therefore, altogether, these data suggest that the vertebrate thalamus, pretectum, and midbrain share a common origin, both ontogenetically (from an early and transient Pax6-positive area found between the prospective ZLI and IsO levels) and phylogenetically (homologous to the amphioxus DiMes region). This hypothesis has two major implications for our understanding of the vertebrate brain Bauplan and its evolutionary origins. First, it implies that two of the diencephalic prosomeres—pretectum (p1) and thalamus (p2)—are more evolutionarily related to the midbrain than they are to the third diencephalic prosomere—the prethalamus (p3)—which would, in turn, be more related with the secondary prosencephalon (see previous section). That is, the diencephalon proper would be neither an evolutionarily nor an ontogenetically primordial subdivision of the vertebrate brain. This striking implication is further supported by the differential responses of these regions to experimental manipulation of the organizers and their associated signaling molecules. Chicken-quail heterotopic grafts of the ZLI, as well as focalized ectopic expression of SHH using beads in chicken embryos, show that only pretectum and midbrain, but not the prethalamus, are competent to be re-patterned to a thalamic fate [6,7,119]. Similarly, quail-chick, rat-chick, or mouse-chick heterotopic grafts of the IsO generate an ectopic midbrain in pretectal and thalamic regions, but never in the prethalamus and secondary prosencephalon [126–129]. That is, thalamus, pretectum, and midbrain have similar developmental potentials that are not shared by the prethalamus. In fact, our hypothesis provides a plausible ontogenetic explanation that has long been missing for these intriguing observations, underscoring its explanatory power.

A second major related implication of our hypothesis is that the vertebrate thalamus, pretectum, and midbrain jointly share altogether a common ancestor with the amphioxus DiMes. Since neither Hh nor Fgf8 and Wnt1, the key morphogens involved in ZLI and IsO activity, respectively, are expressed at the corresponding topological positions in amphioxus [40,69,130–132], it is plausible to speculate that vertebrate thalamus, pretectum, and midbrain partitions may have emerged evolutionarily from an ancestral Pax4/6-positive DiMes-like region concomitantly to the evolution of the ZLI and IsO brain organizers as orthogonal signaling centers. Alternatively, the undivided, small amphioxus DiMes may represent an evolutionary simplification upon the loss of the organizers [47,48], if they were already patterning the neural plate-derived CNS of the last common ancestor of chordates. Irrespectively, a major prediction of both evolutionary hypotheses is that suppression of the organizers during vertebrate development should result in a (relatively) homogeneous, smaller, undivided, and fully Pax6-positive region lying between recognizable prethalamus and hindbrain, as we observed in mouse and zebrafish embryos with suppressed ZLI and/or IsO (Fig 11, and see also [103,107,133–136]). Although the converse experiment—the induction of ectopic organizers in amphioxus—is still not technically possible, future methodological developments could allow assessing if and how the DiMes may respond to these morphogens.

Finally, an independent line of evidence supporting the functional homology between the amphioxus DiMes and the corresponding vertebrate regions comes from the retinal projections in the two lineages. In vertebrates, primary eye projections target mainly the midbrain (optic tectum/superior colliculus), while secondary eye projections target mainly the pretectum and thalamus and, to a lesser extent, prethalamus and hypothalamus [137]. In amphioxus, projections from the single frontal eye have recently been mapped to a Pax4/6-positive region in the four gill slit larval stage [138], which likely corresponds to a DiMes derivative based on its topological position and Pax4/6 expression.

Ethics statement

All animal work in this study has been conducted following the Spanish and European legislation. Adult fish were only used to obtain eggs through natural mating (ethical committee approval number: 635/2014). All mouse experiments were performed according to protocols approved by the Universidad Miguel Hernandez OEP committee (UMH.IN.EP.01.13) and Conselleria Generalitat Valenciana (2014/VSC/PEA/00055). Chicken experiments were performed according to protocols approved by the ethical committee from the University of Murcia (137/2015).

Gene annotation and cloning

For all the previously annotated genes in the B. floridae genome, primer pairs were designed to span the full-length coding sequence when possible. A liquid cDNA library from different developmental stages of the European amphioxus (B. lanceolatum) was screened by PCR using B. floridae specific primers. For previously unannotated genes, we performed tBLASTN searches in the B. floridae JGI v1.0 genome, using the aminoacidic sequences of the vertebrate orthologs. The corresponding genomic sequences were retrieved and a gene model was predicted by GeneWise2 and GeneScan, as previously described [139]. Cloned B. lanceolatum mRNAs used for in situ hybridization are available in S2 Table.

Amphioxus embryo collection, whole-mount in situ hybridization, and histology

Ripe adult amphioxus specimens were collected in Argelès-sur-mer, France. Spawning was induced as previously described [140] in a dry laboratory in Barcelona, Spain. After in vitro fertilization, embryos were cultured at 18 ºC for 15 h or 21 h (4/5 somite and 7 somite stages, respectively) and fixed with 4% PFA in MOPS buffer overnight at 4°C.

Chromogenic whole-mount in situ hybridization was performed as previously described [36] using Nitrobluetetrazolium/bromochloroindolyl phosphate (NBT/BCIP) or BMP purple (Roche) as chromogenic substrate for the final alkaline phosphatase. Following whole-mount in situ hybridization, selected embryos were embedded in a 0.1 M PBS solution with 15% gelatine and 20% sucrose, frozen in isopentane, and sectioned with a cryostat at 12–14 μm-thick. Double-fluorescent in situ hybridizations were performed essentially as nonfluorescent in situ hybridizations, as described in [141] with two extra steps of incubation in 5% NAC and (50 mM DTT, 1% NP40, 0.5% SDS) in PBS1X before the hybridization step.

Dinitrophenol (DNP)-labeled antisense riboprobes were synthesized using DNP-11-UTP labeling reagent (PerkinElmer), and DIG-labeled antisense riboprobes were synthesized using DIG RNA labeling mix (Roche). Labeled riboprobes were detected using anti-DNP-POD (Perkin Elmer) and anti-DIG-POD (Roche) antibodies, and green and red fluorescent signals amplified with TSA -Plus -Fluorescein and Tetrarhodamine systems (Perkin Elmer), respectively.

Images were acquired using a Leica TCS-SPII confocal microscope or a Zeiss Axiophot. Confocal datasets were deconvolved with Huygens Professional version 16.05 (Scientific Volume Imaging, The Netherlands, http://svi.nl), analyzed, and assembled with ImageJ; for panels B and B′ in Fig 8, images were further processed with Imaris (7.2.3, Bitplane AG, software available at http://bitplane.com).

Fish husbandry, morpholino treatments, and in situ hybridization in zebrafish embryos

Breeding zebrafish (Danio rerio) were maintained at 28°C on a 14 h light/10 h dark cycle as described in [142]. To disrupt the ZLI and IsO secondary organizers together, we performed a quadruple transient knockdown using four morpholino-antisense oligomers (MOs) that had been previously described to abolish each of the organizers individually: otx1a and otx2 MO’s for the ZLI [105], and eng2a and eng2b for the IsO [106]. As injection controls, we used a combination of the two nontargeting MOs that were used in the original articles (a morpholino-sense oligomer against twhh (Cont1) [105] and a standard control MO (Cont2) [106]). The combination of experiment or control MOs was injected at the one-cell stage into the yolk at the following concentrations (based on the original sources): otx1a (0.25 mM), otx2 (0.25 mM), eng2a (0.5 mM), eng2b (0.5 mM), Cont1 (0.5 mM), Cont2 (1 mM). Each embryo was injected with 1.5 nl of the MO mix (injection of 1.0 nl produced similar, yet milder, phenotypes, whereas injection of 2.0 nl resulted in full mortality). Four independent experiments were performed (in different days), injecting approximately 100 eggs per condition and experiment.

Injected embryos were fixed in 4% PFA overnight at 4 ºC and used for whole-mount in situ hybridization as previously described [143]. A subset of stained embryos was cryosectioned, and both sections and whole embryos were mounted in 80% glycerol-PBS and photographed in a Zeiss Axiophot microscope. The full list of probe sequences is available in S3 Table.

Analysis of gene knockouts in mice

The Fgf8 conditional mutant was generated by the Gail R. Martin laboratory [120], and the transgenic mouse line expressing cre under the En1 promoter was generated in the Dr. Wolfgang Wurst laboratory [144]. Mutant embryos were generated by crossing double heterozygous males (En1^cre/+; Fgf8^flox/+) with homozygous Fgf8^flox/flox conditional females. Immunohistochemistry (PAX-6) and in situ hybridization (Otx2) in paraffin sections were performed as previously described [145]. The primary PAX-6 rabbit polyclonal IgG antibody was diluted in PBTG (1:500; PRB-278P/Covance). The Otx2 probe was synthesized as in [86].

In situ hybridization in chicken embryos

All the procedures involving extraction of brain samples and further tissue processing were done as previously described [146]. Fertilized chicken (Gallus gallus domesticus) eggs were bought from a national farm (Granja Santa Isabel; Córdoba, Spain) and incubated at 38 ºC and 65% controlled humidity in a forced draft incubator until the Hamburger–Hamilton stage five (HH5) [147]. Embryos were fixed by immersion in 4% paraformaldehyde in 0.1M phosphate buffered saline (PBS, pH 7.4) during 16 h at 4 ºC. Whole-mount in situ hybridization was done as previously described [146] using probes for Otx2 and Gbx2 reported in [24]. Fezf2 probe was cloned using the following primers: F, GCTACAAACCCTTCGTCTGC and R, GCTCAGGGTCACTTGCTACC.