Head regeneration in wild-type hydra requires de novo neurogenesis

Hydra belongs to Cnidaria, a phylum that arose before bilaterians and provides model systems to trace back ancestral developmental processes, as such as apical and anterior patterning(Galliot and Miller, 2000),neuromuscular differentiation(Miljkovic-Licina et al.,2004; Seipel and Schmid,2005) and regeneration(Holstein et al., 2003; Galliot et al., 2006). Hydra polyps display a radial symmetry with an apex or head, centered on a mouth opening surrounded by tentacles that catch and ingest prey, and a basal disk at the opposite extremity. Throughout the animal, the body wall consists of two cell layers, the ectoderm and the endoderm, separated by an extracellular matrix called mesoglea. These two cell layers are made up of myoepithelial cells and interstitial stem cells, which provide precursors for gland cells,neurons, nematocytes and germ cells.

The hydra nervous system is organized as a nerve net that extends throughout the animal and is made up of two cell lineages: the sensory mechanoreceptor cells, named nematocytes, and the neurons, with typical synapses (Westfall, 1996). Those two cell types follow distinct differentiation pathways(Bode, 1996). Nematoblasts undergo several synchronous divisions, forming syncitial cell clusters in the ectoderm of the body column, before differentiating a typical capsule, the nematocyst, and migrating toward the tentacles. By contrast, neuronal precursors follow a more direct differentiation pathway, responding to local cues along the body axis (Fujisawa,1989). Moreover, differentiated neurons constantly change their phenotype as they get displaced toward the extremities(Bode, 1992). Thus, although hydra anatomy appears very simple, highly dynamic processes are required for its maintenance. Besides homeostasis, morphogenesis takes place in sexual and asexual contexts, such as budding, regeneration and reaggregation. Head regeneration, which leads to the replacement of the missing part after 2 days,relies on the setting up of an organizer activity at the regenerating tip(MacWilliams, 1983). This requires the sequential activation of a specific set of `early' genes within the endodermal cells of the regenerating tip, followed 16-20 hours post-amputation (hpa) by the activation of the `early-late' genes at the head patterning stage (Galliot et al.,2006), when an intense cell proliferation takes place in the regenerating tip (Holstein et al.,1991).

According to several independent datasets, neurons are thought to play a minor role in de novo head patterning: chimera experiments demonstrated the primary role of myoepithelial cells in budding rate and regenerative capacity,whereas nerve-free hydra display amazing budding and regenerative abilities(Fujisawa and Sugiyama, 1978; Marcum and Campbell, 1978). More recently the systematic screening of hydra peptides with morphogenetic function identified mostly epitheliopeptides(Fujisawa, 2003). However,neurons were also shown to produce morphogenetic peptides involved in head differentiation (Schaller et al.,1989; Javois and Frazier-Edwards, 1991) and interstitial cells (i-cells) can regulate the morphogenetic potential of the myoepithelial cells(Sugiyama and Wanek, 1993). Therefore, the interplay between interstitial and myoepithelial cells in homeostatic and developmental contexts appears essential but largely unknown.

In this work, we have investigated the putative role of neurogenesis in head regeneration by dissecting the cellular and developmental regulation of cnox-2, the hydra gsx homolog gene. Cnidarian gsx/ind-related genes were identified in numerous cnidarian species,and their expression patterns suggest an ancestral gsx/cnox-2function in neurogenesis and oral patterning. The cnox-2/anthox2genes are activated during apical patterning in hydra(Schummer et al., 1992; Gauchat et al., 2000), display an apical expression in the Hydractinia gastrozooid polyps(Cartwright et al., 1999) and the sea anemone Nematostella juvenile polyps(Finnerty et al., 2003). In the Nematostella planula larva, anthox2 transcripts are localized at the posterior pole, which provides the future oral pole of the polyp. In the early Podocoryne larva, the gsx transcripts,initially localized in the anterior endoderm, extends toward the posterior pole (Yanze et al., 2001). However, in the coral Acropora developing larva, cnox-2expression was detected in neurons of the ectodermal central region of the body axis and rarely in the oral region(Hayward et al., 2001),suggesting some variability in the developmental but not the cellular regulation of the cnidarian gsx/cnox-2 gene family. Nevertheless, in hydra, contradicting data were published as immunochemistry analyses showed a predominant cnox-2 expression in the myoepithelial cells of the body column and a repression during head formation(Shenk et al., 1993a; Shenk et al., 1993b). Therefore a gsx/cnox-2 consensus function remains questionable in cnidarians (Schierwater et al.,2002). We show here that cnox-2 expression in hydra is indeed restricted to the nervous system in both homeostatic and developmental contexts. By analysing the proliferation rate of cnox-2⁺cells, we identified a subpopulation of interstitial stem cells that behave as bipotent neuronal progenitors, giving rise to apical neurons and nematoblasts. Functional assays using RNAi in wild-type hydra indicate that those cnox-2⁺ progenitors promote apical neurogenesis and head patterning during head regeneration.

Hydra vulgaris (Hv, Zürich, Basel, AEP strains), Hydra magnipapillata (Hm-105, sf-1 strains) were cultured as in Gauchat et al. (Gauchat et al.,2004). sf-1 mutants were kept for 2 days at 28°C to eliminate i-cells, and they were subsequently amputated and maintained at 28°C, whereas control sf-1 hydra were kept at 18°C. Bisection was mid-gastric in all regeneration experiments.

Intact hydra (Hm) were continuously incubated with bromodeoxyuridine (BrdU) 5 mmol/l (Sigma) for 2, 24 or 48 hours, immediately fixed, processed for cnox-2 whole-mount in situ hybridization(WM-ISH) and detected for BrdU as in Gauchat et al.(Gauchat et al., 2004). For regeneration experiments, intact hydra were incubated for 4 hours in BrdU,washed, immediately bisected and left in Hydra medium. cnox-2 Hv cDNA was amplified using M13 forward-20 and reverse primers or restricted at the EcoRI site. cnox-2 riboprobes were either DIG- or fluorescein-labeled. Imaging was performed on Axioplan2 and LSM 510 META confocal microscopes (Zeiss).

After hybridization to the mixed DIG-labeled hyZic and fluorescein-labeled cnox-2 riboprobes, the samples were first incubated in the mouse anti-FITC antibody (1:100, Sigma) and detected with the TSA-Kit#2 (Alexa 488, Invitrogen or Molecular Probes). For the second probe a sheep anti-DIG antibody (1:400, Roche) was used, followed by the detection with an anti-sheep HRP (1:100, Sigma) and the TSA-Kit#41 (Alexa 555). The tyramide labeling time was 18 minutes. Samples were washed 3× 10 minutes in PBS, mounted in Mowiol and pictured either at the Zeiss Axioplan2 microscope using a FITC filter or at the Leica SP2 confocal.

Standard immunohistochemistry was performed on either freshly fixed or in situ processed whole-mount animals. Samples were washed in PBS for 3× 10 minutes, treated with 3% H₂O₂/PBS for 1 hour, blocked in 2% BSA and incubated in the mouse monoclonal β-tubulin antibody (1:1500,Sigma clone 2-28-33) ON at 4°C. Subsequently the animals were washed for 3× 10 minutes in PBS and incubated in the anti-mouse HRP antibody(1:100) for 3 hours at room temperature. The samples were detected with the TSA-Kit#2 as above.

The cnox-2 Hv 795 bp PCR product amplified with the cx2-hv-Xmn5′ (GAACCTCTTCTTAAAACGAGTTAG) and the cx2-hv-16HB3(GTAGGGGAATAGCTATATCCTTTCTTA) primers was inserted into the SmaI-digested double T7 vector pPD129.36 (L4440, Fire's laboratory). Double-stranded RNAs (dsRNAs) were produced in HT115(DE3) bacterial strain(Timmons and Fire, 1998). Fifty intact Hv hydra per condition starved for 2 days were given every other day up to nine times grinded agarose that contained either no bacteria, or bacteria having produced control dsRNAs (transformed with the L4440 vector) or dsRNAs corresponding to cnox-2 or Kazal1(Chera et al., 2006). The day following the last dsRNA exposure, samples were processed for RT-PCR or WM-ISH, or bisected for regeneration experiments.

mRNA was prepared from intact hydra (Hv Basel) or upper halves collected immediately post-amputation with the QuickPrep micro-mRNA Purification Kit (Amersham) and resuspended in 10 μl H₂O. mRNA(50 ng) was used for Sensiscript Reverse Transcription (Qiagen) and specific cDNAs were PCR amplified for up to 24 cycles with gene-specific primers(sequences on request).

In three hydra species (Hv, Hm-105, Hvi), cnox-2expressing cells were detected in the ectodermal layer of the adult polyp, in two distinct regions: first in the apex, at the level of the tentacle zone and immediately above; second along the body column but absent from the peduncle and basal disk (Fig. 1A-C,G,H,not shown). Beside single i-cells and pairs of dividing i-cells expressing cnox-2 in both regions, cnox-2⁺ cell types were not identical in these two locations. In the apex, cnox-2⁺cells also displayed a typical neuronal shape, mostly multipolar neurons as evidenced by the β-tubulin co-staining(Fig. 1B,D, and see Fig. S1A-E in the supplementary material), whereas in body column, cnox-2⁺ cells formed clusters, typical of synchronously dividing nematoblasts (Fig. 1C,E). Quantitative analysis of these clusters(Fig. 1F) showed that cnox-2 is expressed at earliest stages of nematoblast proliferation,being detected in pairs of i-cells (20%) and four- and eight-cell clusters(80%), but rarely in 16-cell clusters (<1.3%). Moreover, nematoblasts differentiating a typical capsule never expressed cnox-2(Fig. 1C, asterisks), implying that cnox-2 expression ceases when nematoblasts differentiate.

To verify that cnox-2 expression was restricted to i-cell lineages, we tested the sf-1 mutants that lose their i-cells after a 2 day temperature shift but maintain their myoepithelial cells intact(Marcum et al., 1980). At permissive temperature, cnox-2 expression was as in wild type(Fig. 1G,H), whereas at restrictive temperature cnox-2 expression was abolished(Fig. 1I,J), confirming that cnox-2 expression is absent from the myoepithelial cell lineages. In addition, in wild-type hydra, none of the other i-cell derivatives, including gland cells (not shown), testes and ovaries(Fig. 1K-N) expressed cnox-2. In female hydra, the region facing the ovary was cnox-2 negative (Fig. 1K,L), probably because of the recruitment of i-cells as nurse cells by the mature oocyte. Therefore cnox-2 expression appears restricted to two cell lineages among the i-cells: neuronal at the apical pole and mechanoreceptor along the body column.

Single or twin i-cells form a heterogeneous cell population that includes self-renewing stem cells and precursors to a variety of cell types including gametes, gland cells, neuronal cells and nematoblasts(Holstein and David, 1990a). As cnox-2 transcripts were detected neither in gland cells nor in gametes, the cnox-2⁺ i-cells may represent three distinct populations: self-renewing interstitial stem cells, neuronal precursors and nematoblast precursors. The cycling activity of cnox-2⁺cells was measured through continuous BrdU-labeling(Fig. 2): in body column over 90% cnox-2⁺ cells were BrdU-positive after 2 hours(Fig. 2A,D,G), reaching 100%after 24 hours (Fig. 2B,E,G). Given that S-phase lasts about 12 hours in all i-cells(Campbell and David, 1974) and that a 15 minute incubation is sufficient to detect BrdU-incorporation (data not shown), we calculated that the gastric cnox-2⁺ cells would traverse the cell cycle in about 16 hours if cells are not synchronized.

In the apex, a significant number of cnox-2⁺ cells were BrdU-labeled: about 4, 23 and 45% after 2, 24 and 48 hours, respectively(Fig. 2D-F,H). After 24 and 48 hours' incubation, some of those BrdU⁺/cnox-2⁺cells corresponded to newly differentiated neurons. In fact, pairs of apical BrdU⁺/cnox-2⁺ cells were often observed,showing extended processes, corresponding to precursors that differentiate into cnox-2 neurons immediately after mitosis(Fig. 2E). Thus cnox-2⁺ i-cells correspond to a subset of cycling i-cells that provide apical neurons and proliferating nematoblasts and can therefore be regarded as bipotent neuronal progenitors.

Hydra starvation leads to dramatic cellular rearrangements in both epithelial (Bosch and David,1984; Holstein et al.,1991) and interstitial(Gauchat et al., 2004) cell compartments. In case of cnox-2⁺ cell clusters, their distribution was not significantly affected over the first 3 days, except for a slight decrease in the single i-cell compartment(Fig. 1G). However, when starvation was prolonged over 10 days, a drastic reduction in the number of gastric cnox-2⁺ cells was recorded(Fig. 3G), whereas the number of apical cnox-2⁺ cells did not vary(Fig. 3F). Hence starvation drastically alters the production and/or survival of cnox-2⁺ nematoblasts, but not of the apical cnox-2⁺ lineage.

We next tested cnox-2 function through RNAi loss-of-function assay. Kazal1 was selected as control gene, as its expression is restricted to a distinct i-cell lineage, the gland cells(Chera et al., 2006). Moreover, cnox-2 and Kazal1 did not show any epistatic relationships,as evidenced by their conserved respective expressions in Kazal1(-) and cnox-2(-) hydra(Fig. 3B, Fig. 4A,B). cnox-2silencing was effective and specific, as shown by RT-PCR and WM-ISH assays:hydra exposed repeatedly to cnox-2 dsRNAs exhibited a drastic reduction in cnox-2 transcripts levels, whereas actin,Kazal1 and CREB levels remained unaffected(Fig. 3A, Fig. 4B). When whole hydra were compared to upper halves (enriched in apical tissue), cnox-2silencing appeared stronger in the former ones, even though the control level of cnox-2 expression was significantly lower in whole hydra than in apically enriched tissue (Fig. 3A, lanes 1,3). Therefore, the decrease in cnox-2 levels in the body column probably corresponded to the addition of two distinct effects: the specific cnox-2 RNAi knockdown and the starvation effect described above, leading to a drastic decrease in nematoblast production. WM-ISH confirmed the progressive disappearance of cnox-2⁺cells with the successive dsRNA exposures, becoming very rare after five feedings in gastric (Fig. 3C)and apical (Fig. 3D,E) regions. The number of cnox-2⁺ apical cells was significantly reduced after five exposures, by 55 and 38% when compared to mock-silenced and Kazal1(-) hydra, respectively(Fig. 3F). In body column, a single dsRNA exposure led to a 48% decrease in cnox-2⁺cell number (Fig. 3G, lanes 1-3). However, after five exposures, the 12 day starvation effect was dramatic, and the difference between control and cnox-2 RNAi hydra was no longer significant (Fig. 3G, lanes 4-6). This loss of function obviously affected hydra homeostasis, as cnox-2(-) hydra progressively became significantly smaller than Kazal1(-) or control hydra(Fig. 3B,C): about twice as short as control hydra after nine exposures (hydra exposed 9× to cnox-2 dsRNAs: 0.44 mm ± 0.18; to control dsRNAs: 0.85 mm± 0.28, n=10).

To detect whether the decrease in cnox-2⁺ apical expression affected the organization of the apical nervous system (ANS)(Fig. 3H-M), we used the anti-β-tubulin antibody (Siddiqui et al., 1989) that stains all types of neurons, nematocytes and some i-cells, but not nematoblasts or myoepithelial cells(Fig. 1D,E, and see Fig. S1G,H in the supplementary material). This staining showed a highly organized ANS,with parallel processes of the sensory neurons along the hypostome (the domed region from the mouth opening down to the tentacle zone) providing an inverted-basket aspect (Fig. 3H,I, arrows, and see Fig. S2 and Movie S1 in the supplementary material). On sagittal views, the multipolar neurons were visible, spread along meridian lines that converge toward the mouth opening(Fig. 3H). After five cnox-2 dsRNA exposures, the ANS was dramatically reduced, restricted to its most apical part, whereas the sub-tentacle zone became nerve-free. The density of sensory neurons was decreased, the systematic parallel orientation of their processes was altered and their connections to the multipolar neurons were no longer visible (Fig. 3J,L, and see Fig. S2 in the supplementary material). After nine exposures, apical neurons were no longer detected(Fig. 3N), indicating that cnox-2 expression is required to maintain apical neurogenesis and ANS organization.

Previous cellular expression analyses identified several candidate genes as regulators of the nematocyte differentiation pathway. Those genes display cellular expression patterns with striking differences when considering the proliferation rate, the cluster size and the presence of a differentiating capsule. cnox-2 is expressed in precursors and at early proliferative stages but is repressed as soon as nematoblasts differentiate (this work); hyZic is also expressed at early stages(Lindgens et al., 2004), but with a lower representation among precursors (6% for hyZic, over 20%for cnox-2) and a lower BrdU-labeling index after a short incubation than cnox-2 (40 versus 95%). Therefore, hyZic is probably turned on at a slightly later stage than cnox-2 and appears to be a candidate cnox-2 target gene. As additional regulators, the hyCOUP-TF orphan receptor was proposed to repress proliferation in nematoblast clusters, switching them to differentiation(Gauchat et al., 2004),whereas the achaete-scute homolog CnASH is expressed in non-proliferative differentiating nematocytes(Lindgens et al., 2004). Finally, the transcription factor CREB is expressed at all proliferative stages of this pathway (Chera et al.,2007).

Cellular expression analysis showed that cnox-2 and hyZictranscripts indeed colocalize in most i-cells and nematoblast clusters(Fig. 4A). Moreover, in cnox-2(-) hydra, hyZic and CnASHtranscripts were undetectable, hyCOUP-TF was upregulated and CREB was unaffected (Fig. 4B, left). These results support an upstream role for cnox-2 (Fig. 4C).

As well as the nematocyte pathway, hyCOUP-TF, CnASH and CREB are also expressed in the neuronal lineage(Gauchat et al., 2004; Hayakawa et al., 2004; Chera et al., 2007), whereas other genes display a neuronal-specific expression, such as the PRD-class prdl-a (Gauchat et al.,1998) and gsc (Broun et al., 1999), the ANTP-class msh(Miljkovic-Licina et al.,2004) and the RFamide neuropeptide genes(Mitgutsch et al., 1999). Their RT-PCR expression analysis in cnox-2(-) and control hydra showed a significant downregulation of prdl-a, gsc and RFamide-B, whereas hyCOUP-TF was found to be upregulated(Fig. 4B, right), indicating that cnox-2 contributes to the regulation of those neuronal apical genes. Moreover, msh expression in the gastric region was slightly decreased, suggesting that gastric cnox-2⁺ i-cells also provide neuronal precursors in this region.

De novo neurogenesis, restricted to head-regenerating tips and occurring at a time depending on the bisection level, early after decapitation and`early-late' after gastric bisection, was previously reported(Venugopal and David, 1981). To monitor the cellular remodeling occurring in head-regenerating halves, we used the anti-β-tubulin staining and observed the complete disappearance of any neurons from head-regenerating tips for the first 16 hpa(Fig. 5A, and see Fig. S3 in the supplementary material). Thereafter, pairs of i-cells appeared(Fig. 5A, arrows), becoming more and more numerous, until a neuronal network was detected again when tentacle buds (TBs) emerged.

Interestingly, in head-regenerating tips, cnox-2 displayed two opposite successive regulations that parallel the de novo neurogenesis described above. For the first 20 hpa, the head-regenerating tips were devoid of cnox-2⁺ cells (Fig. 5B). Subsequently dividing cnox-2⁺ i-cells first became visible (Fig. 5B,D, Fig. 6A),preceding the appearance of neuronal precursors and differentiated cnox-2⁺ neurons (Fig. 5E,F, Fig. 6E),which gradually reached a maximum density at 48 hpa(Fig. 5J). This early-late cnox-2 regulation, in agreement with the previously reported increase in cnox-2 transcripts in the head-regenerating H. viridissima at 24 hpa (Schummer et al., 1992), was specific to the head-regeneration process, as it was not detected in foot-regenerating tips whatever the time point(Fig. 5B,C, arrowheads). By contrast, the cnox-2⁺ neuronal density in the heads of foot-regenerating halves remained stable all through the regeneration process(Fig. 5C, right panel).

Along the body column a simultaneous decrease in the number of cnox-2⁺ cells in both head- and foot-regenerating halves was observed at 24 and 36 hpa, lowered by 40-50% of the normal level in undisturbed polyps (Fig. 5C,H). This downregulation affected proliferating nematoblasts but not single i-cells, those actually showing a 2-fold increase of their density at 16 hpa(Fig. 5I). Therefore this reduction in gastric cnox-2⁺ cells probably corresponds to the cell death of differentiating nematoblasts linked to the amputation stress as previously reported (Fujisawa and David, 1984), a scenario supported by the transient relative increase in single cnox-2⁺ i-cells.

During budding, numerous cnox-2 cells were detected in the distal tip from stage 3 onward (Fig. 5G). This stage, which occurs 16-20 hours before TBs become visible (Otto and Campbell,1977), corresponds to the onset of the early-late transition during regeneration. Similarly to head-regenerating tips, those cnox-2⁺ cells were identified as dividing i-cells,neuronal precursors and differentiated neurons, first broadly distributed,becoming restricted to the tentacle zone and the basis of the hypostome at stage 7-8. Therefore, during budding and regeneration, cnox-2exhibits a similar spatiotemporal regulation in the presumptive head region;the initiation of its expression in the neuronal cell lineage precedes by 20 hours the emergence of the TBs. However, during budding, cnox-2⁺ expression in the body column of the growing bud was not altered (Fig. 5G),indicating that the negative `post-amputation' gastric regulation does not take place here.

To trace back the origin of the cnox-2⁺ neurons detected in head-regeneration tips, hydra were BrdU-labeled for 4 hours before bisection and BrdU⁺/cnox-2⁺ cells were analysed during early-late regeneration (Fig. 6). From 20 to 44 hpa, the density of BrdU⁺ cells progressively increased in regenerating tips(Fig. 6A-C), including BrdU⁺/cnox-2⁺ cells surrounding the newly formed mouth opening (Fig. 6B,C). Interestingly most cnox-2⁺ cells were BrdU⁺, corresponding to pairs of i-cells(Fig. 6D,F) and differentiated neurons (Fig. 6E). These pairs of i-cells, either asymmetric (Fig. 6D) or symmetric (Fig. 6F), frequently exhibited elongated processes, indicating that neuronal differentiation was taking place, resulting in mature neurons that were rare at 20 hpa but frequent at 44 hpa, still exhibiting some BrdU labeling. In a few BrdU⁺ pairs of differentiating neurons, one of the two cells did not express cnox-2 (not shown). These data indicate that the cnox-2⁺ neurons detected at the early-late phase of head regeneration result from a differentiation process of cnox-2⁺ neuronal precursors, rather than induction of cnox-2 expression in mature neurons that would migrate and express cnox-2 once they have reached the regenerating tip.

sf-1 hydra maintained at a permissive temperature regenerated their head efficiently, although delayed by 60 hours when compared with wild-type hydra (Fig. 7A). At a restrictive temperature, regeneration was poorly efficient, as 45% hydra had not regenerated their head after 6 days and subsequently died(Fig. 7A). To test a possible correlation between the level of cnox-2 expression and the efficiency of the head regeneration process, hydra were collected at different time points and sorted out according to both their cnox-2⁺apical cell number (null, low: fewer than 20, normal: more than 20) and their phenotype (Fig. 7B). The phenotypes were identified as Stage 1, `ball-shape' with no obvious apical pole; Stage 2, `elongated' with clearly distinguishable basal and apical poles; Stage 3, `TB1' when the TBs have just emerged; Stage 4, `TB2' when more than two TBs elongate; Stage 5, `fully head-regenerated', similar to the adult head (Fig. 7C-H).

The ball-shaped hydra (n=22) never showed any cnox-2expression (Fig. 7C); this was also the case for most elongated hydra (75%, n=17, Fig. 7D), the others displaying a few cnox-2⁺ cells (not shown). By contrast, most TB1 hydra exhibited a few cnox-2⁺ cells (82%, Fig. 7E, arrows, n=16), the others (18%) being null for cnox-2. Among the latter, some probably underwent an `aborted' regeneration process, as evidenced by their abnormal shape and the presence of a single tentacle bud(Fig. 7G,H, arrowheads). The emergence of a single tentacle was never observed in control hydra. At Stage 4(n=31) and Stage 5 (n=14), 45 and 57% hydra displayed a normal number of cnox-2⁺ cells, respectively(Fig. 7F); nevertheless a significant number of hydra exhibited no (16 and 7%) or low (39 and 29%) cnox-2 expression. We also noted some sporadic gastric single cnox-2⁺ cells but never observed cnox-2⁺ cell clusters.

By contrast, the cnox-2 expression pattern in regenerating sf-1 hydra maintained at a permissive temperature was similar to that recorded in wild-type hydra: numerous cnox-2 apical cells at the elongated shape stage (Fig. 7I)and a dense cnox-2⁺ neuronal network at later stages(Fig. 7J,K). Along the body column, cnox-2⁺ cell clusters became visible first in the aboral region (Fig. 7I) and then homogenously distributed along the axis(Fig. 7K). These data suggest a correlation between the level of cnox-2 apical expression and the efficiency of the head-regeneration process, even though in a few cases (7%),a complete head regeneration process occurred when cnox-2 expression was lacking or only transient.

To test the cnox-2 function during head regeneration, hydra were exposed to dsRNAs repeatedly and bisected. Emergence of TBs was delayed by 42 hours in those hydra when compared with controls and at 39 hpa, the ANS was not formed (Fig. 8A,I,J). The number of cnox-2⁺ apical cells detected in head-regenerating tips of Kazal1(-) hydra increased over time in a similar way to non-treated hydra(Fig. 8B,D,F). By contrast, at 30 hpa twice as few cnox-2⁺ apical cells were detected in cnox-2(-) hydra (Fig. 8E, arrow) as in Kazal1(-) or control hydra (55 and 53%, respectively, Fig. 8B). At 48 hpa, two distinct phenotypes were observed: strong(Fig. 8G) when the animal size was small, the regeneration stacked as evidenced by the absence of TBs(arrowheads in Fig. 8F) and the number of cnox-2⁺ apical cells low(Fig. 8B); weak when the animal size was larger (although smaller than in Kazal1(-) hydra),the appearance of TBs only delayed (Fig. 8H, arrowhead) and the number of cnox-2⁺neurons closer to that observed in control hydra (35 per tip, Fig. 8B, black bar). Thus a clear correlation between the level of cnox-2⁺ apical expression and the efficiency in de novo head formation was observed in cnox-2 dsRNA-treated hydra.

Interestingly, when considering the number of gastric cnox-2⁺ cell clusters, no correlation with the head-regeneration phenotype was evidenced. In fact, at 30 hpa, this number was similar in Kazal1(-) and cnox-2 knockdown hydra. However, at 48 hpa, cnox-2(-) hydra had not yet reestablished cnox-2 expression in their body column(Fig. 8C,G,H), by contrast to control and Kazal1(-) hydra(Fig. 8F).

In this work we show that cnox-2⁺ cells correspond to precursors as well as derivatives of two distinct cell lineages, taking into account their position along the body axis, their morphology and their clustered organization. These are: (1) a subset of neurons and their precursor cells in the apex; and (2) dividing i-cells as well as proliferating nematoblasts in the body column. This fate restriction of cnox-2⁺ cells is supported by the complete and rapid disappearance of cnox-2⁺ cells in sf-1 mutants. Hence these two cell lineages, which are known to share a common interstitial stem cell (Bode, 1996), also share a common set of evolutionarily-conserved regulatory genes, which, in addition to the ParaHox gsx/cnox-2 gene (this work), includes the hyCOUP-TF nuclear orphan receptor, the PRD-CLASS homeogene prdl-b (Gauchat et al.,2004) and the Achaete-scute homolog CnASH(Hayakawa et al., 2004). None of these four genes were found expressed in any other derivatives of the interstitial stem cells or in myoepithelial cells. Among the hyCOUP-TF⁺ and prdl-b⁺ cells, single i-cells were not observed and pairs of dividing i-cells were rare(Gauchat et al., 2004). Furthermore, CnASH is a marker for non-proliferating differentiating nematoblasts (Lindgens et al.,2004) and differentiating sensory neurons(Hayakawa et al., 2004). Therefore, cnox-2 occupies a unique position among the putative regulators of the hydra nervous system, as it identifies a highly proliferative subset of i-cells, the progeny of which are restricted to neuronal and mechanosensory cell fates. Similarly, a subpopulation of i-cells was shown to be restricted to a gametic fate(Littefield, 1991; Nishimiya-Fujisawa and Sugiyama,1993), but this is the first report concerning the somatic lineages. Previous studies showed that neuronal precursors from the peduncle could actually provide nematocyte precursors after transplantation into a gastric environment (Holstein and David,1990b), suggesting that these two cell lineages indeed share common specific precursors. In the absence of one of the criteria that define stem cells, i.e. the evidence that these cells can self-renew(Weissman et al., 2001), we propose to consider the cnox-2⁺ i-cells as progenitors to neurons and mechanosensory cells (Fig. 4C). Interestingly, in mice, the gsx homologs Gsh1 and Gsh2 appear to control the size and the identity of neuronal progenitor pools (Toresson and Campbell, 2001; Yun et al.,2003).

Although genes expressed in neurons are readily refractory to RNAi(Tavernarakis et al., 2000),the feeding RNAi strategy we applied for the first time in hydra to silence a neurally expressed gene, indeed induced specific effects, including a drastic disorganization of the ANS and a deficient head-regeneration process. In both contexts, intact or regenerating hydra, the number of cnox-2⁺ cells (i-cells, apical neurons, nematoblasts) was significantly reduced after repeated exposures to cnox-2 dsRNA. Moreover, the prdl-a, gsc, RFamide-B and RFamide-C that are specifically expressed in apical neurons, were strongly downregulated. Interestingly, the role of cnox-2 in neurogenesis also probably extends to the body column, as the expression of the non-apical neuronal gene msh was altered in cnox-2(-) hydra. However, in the body column, we did not identify cnox-2+ neurons, suggesting that in this region cnox-2 function is restricted to neuronal progenitors,whereas in the apex cnox-2 would be required at least at two distinct levels, to commit interstitial cells to the neuronal fate and to promote the differentiation of multipolar neurons.

The striking genetic regulations observed in cnox-2(-)hydra also indicate possible epistatic relationships in the nematocyte pathway(Fig. 4C). cnox-2appears as one of the earliest activated genes and possibly regulates the commitment of i-cells to the nematocyte pathway as well as the proliferation of nematoblasts, turning on hyZic, directly or indirectly. By contrast, cnox-2 and/or hyZic probably negatively regulate hyCOUP-TF in proliferating nematoblasts, because hyCOUP-TFis supposed to act as a repressor of proliferation(Gauchat et al., 2004) and is upregulated in cnox-2(-) hydra. As previously proposed, hyZic might regulate CnASH positively in differentiating nematocytes (Lindgens et al.,2004). At least two genes expressed in this pathway were not affected: CREB and NOWA, a structural gene involved in nematocyst differentiation (Engel et al.,2002). CREB could act upstream of cnox-2 in this cell lineage. However, as CREB is strongly expressed in i-cells, proliferating nematoblasts, differentiated neurons and myoepithelial cell lineages(Kaloulis et al., 2004; Chera et al., 2007), its regulation within a specific cell lineage cannot be uncovered in RT-PCR assays and requires further analyses. Finally the stable low level of NOWAtranscripts suggests that a longer period of cnox-2 silencing might be required before differentiated nematocytes are affected.

Genes expressed within the head-regenerating tips exhibit modulations that are temporally characterized as `immediate', `early', `early-late' and `late',matching with the wound healing, the setting up of the organizer activity, the proliferation phase or following head formation, respectively(Galliot and Schmid, 2002). As cnox-2 expression in the presumptive head region resumes after 16 hpa, cnox-2 belongs to the `early-late' genes. Previous hydroxurea experiments showed that cells in S-phase at the time of amputation provide newly differentiated neurons in head-regenerating tips(Venugopal and David, 1981; Holstein et al., 1986). The analysis of the BrdU⁺/cnox-2⁺ cells confirms this result: i-cells in S-phase at amputation time undergo mitosis and neuronal differentiation 20 to 30 hours later in head-regenerating tips. As those tips contain very few BrdU⁺ cells immediately after amputation (Holstein et al.,1991) (S.C. and L.G., unpublished), we anticipate that those cells first migrate toward the tip and become cnox-2⁺ upon exposure to head-specific differentiation signal(s). This scenario fits with the injury effects previously described: injury promotes the migration of i-cells (Fujisawa et al.,1990), which is enhanced by the signals released by the head-regenerating tips (Teragawa and Bode,1991). Those signals also drive the terminal differentiation of neuronal precursors within about 18 hours(Venugopal and David, 1981; Holstein et al., 1986). Hence, cnox-2 appears to be a target gene for neurogenic signals during head formation. A similar continuous phenotypic maturation of neurons occurs from the gastric region to the basal extremity, accelerated upon amputation(Technau and Holstein, 1996),suggesting that de novo neurogenesis rather than phenotypic conversion of existing neurons (Bode, 1992)is the predominant mechanism during regeneration.

In two distinct contexts in which cnox-2 expression was reduced or undetectable, i.e. RNAi-silenced wild-type hydra and sf-1 mutants,head regeneration was dramatically altered. As cnox-2 supports proliferation of i-cells and de novo neurogenesis at the early-late stage,these results support the importance of these two cellular processes in head-regenerating tips for head patterning, and the possible contribution of neuronal cells as previously proposed(Schaller et al., 1989).

Nevertheless, in a few sf-1 hydra in which the number of cnox-2⁺ cells was low or null, the head regeneration process was advanced or complete. We anticipate that such animals maintained for several days at restrictive temperature might have transiently expressed cnox-2, allowing the head formation process to proceed, subsequently losing cnox-2 expression and/or cnox-2⁺ cells. Alternatively a nerve-independent regeneration process might have taken over in these animals. Similarly, in cnox-2(-) hydra that had ultimately regenerated their head, either cnox-2 silencing was transient and cnox-2 expression reestablished at the time of head formation, or a cnox-2-independent alternative mechanism was activated. Recently, a similar correlation between the disappearance of Dickkopf+ gland cells and the blockade in head regeneration in sf-1 hydra was reported (Guder et al., 2006), suggesting that gland cells, besides their immediate cytoprotective effect (Chera et al.,2006), also participate in the head regeneration process.

Comparative analysis of the regeneration processes in urodeles and hydra imposed opposed views concerning their respective nerve-dependence: in urodeles nerve-dependence is complete, as neurotrophic factors are required for the proliferation of blastema cells(Singer, 1974), whereas in hydra, nerve-dependence is dispensable, as nerve-free animals regenerate(Marcum and Campbell, 1978; Sugiyama and Fujisawa, 1978). However, in urodeles the nerve-dependence is linked to the preexisting homeostatic conditions, as limbs that developed in the absence of nerves(aneurogenic limbs) are able to regenerate in the absence of any neuronal support (Brockes, 1987; Tassava and Olsen-Winner,2003), mimicking the nerve-free hydra situation. In such hydra,the genetic program at work in epithelial cells is not yet known, but is likely to be different from that at work in wild-type hydra(Schaller et al., 1980; Hornberger and Hassel, 1997)(S.C., unpublished). In the head-regeneration deficient mutant reg-16, removal of i-cells rescues head regeneration, highlighting the morphogenetic consequences of a misregulation between i-cells and myoepithelial cells (Sugiyama and Wanek,1993). We propose that in the wild-type context, at the stage of head formation, hydra makes use of cell proliferation and de novo neurogenesis, both requiring cnox-2 activity. This sequence of events would be the most efficient and the fastest way to achieve head regeneration. In the absence of one of these components, i-cell proliferation, neuronal differentiation and cnox-2 activation, other routes can be taken,although much slower and much less efficient.

We thank Virginie Voeffray and Fabienne Chabaud for excellent technical help, Corina Guder and Thomas Holstein for providing the sf-1 hydra,Alejandro Sanchez-Alvarado and Jeremy Brockes for helpful discussions, and Ariel Ruiz-i-Altaba and Volker Schmid for precious comments on this manuscript. This work was supported by the Swiss National Foundation, the Geneva Canton, the Fonds Claraz and the Geneva Academic Society.