Maternal and Zygotic aldh1a2 Activity Is Required for Pancreas Development in Zebrafish

We have isolated and characterized a novel zebrafish pancreas mutant. Mutant embryos lack expression of isl1 and sst in the endocrine pancreas, but retain isl1 expression in the CNS. Non-endocrine endodermal gene expression is less affected in the mutant, with varying degrees of residual expression observed for pdx1, carbA, hhex, prox1, sid4, transferrin and ifabp. In addition, mutant embryos display a swollen pericardium and lack fin buds. Genetic mapping revealed a mutation resulting in a glycine to arginine change in the catalytic domain of the aldh1a2 gene, which is required for the production of retinoic acid from vitamin A. Comparison of our mutant (aldh1a2um22) to neckless (aldh1a2i26), a previously identified aldh1a2 mutant, revealed similarities in residual endodermal gene expression. In contrast, treatment with DEAB (diethylaminobenzaldehyde), a competitive reversible inhibitor of Aldh enzymes, produces a more severe phenotype with complete loss of endodermal gene expression, indicating that a source of Aldh activity persists in both mutants. We find that mRNA from the aldh1a2um22 mutant allele is inactive, indicating that it represents a null allele. Instead, the residual Aldh activity is likely due to maternal aldh1a2, since we find that translation-blocking, but not splice-blocking, aldh1a2 morpholinos produce a phenotype similar to DEAB treatment. We conclude that Aldh1a2 is the primary Aldh acting during pancreas development and that maternal Aldh1a2 activity persists in aldh1a2um22 and aldh1a2i26 mutant embryos.


Introduction
Similar to the pancreas of other vertebrates, the zebrafish pancreas consists of an endocrine and an exocrine portion. The zebrafish exocrine pancreas consists of acinar cells that release digestive enzymes into the intestine and the endocrine pancreas is composed of five cell types that secrete hormones directly into the blood stream; insulin producing b-cells, somatostatin producing dcells, glucagon producing a-cells, pancreatic polypeptide hormone secreting PP-cells and ghrelin producing e-cells [1,2]. The zebrafish pancreas develops from a dorsal and a ventral bud associated with the gut tube, where the dorsal bud is located slightly posterior to the ventral bud [3,4]. The dorsal bud is the first to form at 24 hpf and eventually gives rise to endocrine pancreas. By 40 hpf, the ventral bud has formed and is composed of exocrine cells as well as a few endocrine cells. By 52 hpf, the two buds have merged to form one organ on the right side of the embryo, consisting of a single islet of endocrine cells surrounded by the exocrine pancreas [3,4].
RA is involved in the formation of the central nervous system, lung, kidney, intestine, and pancreas [12,15,17,18,19,20]. In particular, RA is needed at the end of gastrulation for pancreas development and blocking RA signaling in zebrafish embryos prevents pancreas formation [15]. Accordingly, exogenously applied RA induces ectopic pancreatic gene expression in the anterior endoderm [15]. Experiments in amphibian and avian models give similar results, indicating a vertebrate requirement for RA in pancreas development [7,12,13]. RA is a small lipophilic molecule derived from dietary vitamin A (retinol). Retinol is converted to an aldehyde (retinaldehyde) which is further converted to a carboxylic acid (retinoic acid). The first step, oxidation of retinol to retinaldehyde, is made possible by several retinol dehydrogenases (RDHs) that have widespread and overlapping expression patterns. The second step, oxidation of retinaldehyde to RA, is carried out by retinaldehyde dehydrogenases (Raldh or Aldh), which have more tissue specific expression patterns [21,22,23,24,25]. In particular, aldh1a2 (raldh2) is the major retinoic acid generating enzyme in the early mouse embryo and was thought until recently to be the only raldh expressed in zebrafish. Recently, aldh1a3 (raldh3) and aldh8a1 (raldh4) were identified in zebrafish [26,27] but aldh1a1 (raldh1) has not been found in zebrafish to date. aldh1a3 is expressed in the developing eye and ear after gastrulation and aldh8a1 is expressed later around 2 dpf in the liver and intestine [26,27] suggesting that these genes are not involved in early pancreas development. In contrast, aldh1a2 is expressed at 30% epiboly in the mesendoderm and continues to be expressed in the posterior and lateral mesoderm during segmentation [28]. At later stages, aldh1a2 is expressed in the somites and the pronephric anlage (by 15 hpf) as well as in pharyngeal arch and pectoral fin mesenchyme (32 hpf) [28,29,30,31,32,33]. Expression of aldh1a2 adjacent to, but not within, the pancreatic anlage is consistent with observations that the anterior paraxial mesoderm is a source of RA driving pancreas formation. Accordingly, three Retinoic Acid Receptors (two RARa and one RARc) are expressed in the endoderm, indicating that the RA signal can be received directly in the endoderm [8].
We carried out a haploid ENU (N-ethyl-N-nitrosourea) screen for endocrine pancreas mutations and discovered a mutant (88.21) that does not develop isl1 expression in the endocrine pancreas, but maintains isl1 expression in the CNS. More detailed analysis of the 88.21 mutant revealed residual expression of several pancreas (e.g. pdx1) and liver (e.g. hhex and prox1) genes, suggesting that endoderm organ differentiation, including pancreas formation, is not completely lost in the mutant. We mapped the 88.21 mutant using a CA panel and identified a mutation in the catalytic domain of the aldh1a2 gene; therefore we named our mutant aldh1a2 um22 . Two other mutant alleles for aldh1a2 have been reported, neckless (nls or aldh1a2 i26/i26 ; a point mutation in the NAD binding domain) and no fin (nof or aldh1a2 u11/u11 ; a point mutation in the catalytic domain) [28,34]. A detailed analysis of endoderm gene expression in aldh1a2 i26 embryos revealed residual expression of several endoderm markers, (e.g. pdx1), similar to the phenotype seen in aldh1a2 um22 mutants. In contrast, we find that embryos treated with DEAB (diethylaminobenzaldehyde), a competitive reversible inhibitor of all Aldhs, completely lack expression of all pancreas and liver genes, indicating that there is residual Aldh activity in aldh1a2 um22 and aldh1a2 i26 mutant embryos. Notably, targeting both maternal and zygotic transcripts using MOs to the aldh1a2 translation start site produces a phenotype comparable to DEAB treatment. In contrast, targeting primarily zygotic transcripts using MOs to the exon1/intron1 splice site of aldh1a2 does not fully block endodermal gene expression. Our results reveal an absolute requirement for Aldh activity in pancreas development and demonstrate residual Aldh activity in aldh1a2 um22 and aldh1a2 i26 mutants, likely due to maternally contributed Aldh1a2.

Fish Maintenance
Ekkwill (EK), Tupfel long fin (TL) and neckless (aldh1a2 i26 ) (Gift from Prince Lab) embryos were collected from natural matings and reared in 1/3 Ringer's. Embryos were staged using morphological criteria up to 24 hours post fertilization (hpf) and then by time of development at 28.5uC [35].

ENU Screen
EK males were treated with 3 mM ENU (N-ethyl-N-nitrosourea) once a week for 3 weeks. The males were then crossed repeatedly to clean out any post meiotic germ cells that were mutagenized. Mutagenized males were then crossed to EK females and the progeny (F1) were raised. Haploid embryos were produced by In Vitro Fertilization (IVF) of F1 female progeny with irradiated sperm. Haploid embryos were raised to approx-imately 30 hpf and fixed in 4% paraformaldehyde for in situ hybridization with islet1 (isl1) probe. Embryos were screened based on isl1 expression. F1 females that produced embryos with mutant phenotypes were out-crossed to TL males and the progeny (F2) were raised and in-crossed for recovery of mutation in diploid embryos.

Mapping, DNA Extraction, RNA Extraction and cDNA Synthesis
Mutant carriers were in-crossed and progeny raised to 4 dpf. Embryos were sorted based on their phenotype; mutants develop a swollen pericardium and lack fin buds. Genomic DNA was extracted from phenotypically mutant and phenotypically wild type embryos at day 4. DNA pools were created from phenotypically mutant and wild type embryos. Bulk segregant analysis was performed on the DNA pools using a 192 CA marker panel [36,37,38]. Two markers were found to be linked to the mutation: z10441 (FW:GCATTCAGATTCTGGGGTGT, RV: CGGAT-GAACCCATCAATCTC) and z8693 (FW: GCTTTTTGAGCA-GATGAGGC, RV: CATGTACGCGTTGACTTTGC). PCR was performed on individual embryos using the same primers. cDNA was synthesized from RNA extracted from pools of 10 phenotypically mutant and 10 phenotypically wild type embryos using Invitrogen Superscript III Reverse Transcriptase Kit. PCR primers, FW: CCAAAGTTGTAATCGCACATC, RV:TTTT-TTTTTTTTTTTCAGAGGTAAAAC, were used to clone fulllength aldh1a2 cDNA. Stratagene Hi Fi taq polymerase was used in the PCR and the product was sequenced. Primers FW: AGCGGCCGTCTTCCCAGAGATATC and RV: GGAATG-GGTGTAGGCAGTTAATGGTGG were used to sequence aldh1a2 from individual embryos.

mRNA and Morpholino Injections
An antisense morpholino oligo (MO) designed to block translation of the aldh1a2 mRNA (tMO) 59GCAGTTCAACTT-CACTGGAGGTCAT39 [28] and one control mismatch morpholino (mmMO 59GCAcTTgAACTTCAgTGGAcGTgAT39 that has five mismatches relative to tMO) were obtained from Gene Tools. 1 nl of 100 uM, 250 uM, 500 uM, 750 uM and 950 uM of tMO was injected at the 1-2 cell stage. A splice MO (sMO) designed to exon1/intron1 splice junctions: 59TTGAAAAAGTCCGA-CAAACCTTGGT39 and one control morpholino (mmMO: 59TTcAAAAAcTCgGACAAtCCTTcGT39 with five mismatches relative to sMO) was obtained from Gene tools. 1 nl of 500 uM, 750 uM, and 950 uM of sMO was injected at the 1-2 cell stage.
For rescue experiments, the aldh1a2 ORF was amplified from TL embryos or aldh1a2 um22 mutant embryos using FW: AT-GACCTCCAGTGAAGTTGAACTGCCA and RV:TTAAGA-CGTCTTGCTTCATCGTAATGGTTTTCA. Both ORFs were cloned with Invitrogen Topo TA Cloning kit, digested using EcoR1 and cloned into PCS2+. Constructs were linearized with Not1 and Ambion Kit Sp6 was used to make mRNA. 500 pg of mRNA was injected into an aldh1a2 um22 in-cross at 1-2 cell stage. mRNA and MO injected embryos were fixed in 4% paraformaldehyde at various developmental stages for in situ hybridization.

DEAB Treatment
A 1 mM stock of DEAB was dissolved in DMSO. Embryos were treated in the dark with 1 uM, 5 uM and 10 uM of DEAB dissolved in 1X PTU at 6, 8, 10, or 12 hpf. Embryos were fixed at various stages and assayed by in situ hybridization. Control embryos were treated in DMSO under similar conditions.

An ENU Screen for Zebrafish Pancreas Mutants
A haploid in situ hybridization screen of ENU (N-ethyl-Nnitrosourea) mutagenized zebrafish was carried out to identify mutations in endocrine pancreas development. Ekkwill (EK) males were treated with 3 mM ENU and crossed to EK females. F1 progeny was raised and eggs from F1 females were in vitro fertilized using irradiated sperm from EK males. The resulting haploid embryos were raised until 30 hpf and assayed by in situ hybridization for islet1 (isl1) expression to detect defects in the endocrine pancreas. F1 females that produced clutches with 50% mutant embryos were outcrossed to Tupfel long fin (TL) males. F2 progeny were raised and screened for recovery of the mutation in the F3 generation. We screened 200 genomes and discovered ten females with defective endocrine pancreas formation. Six of the ten females died, developed tumors or did not produce progeny. Out of the remaining four females, we recovered diploid mutants for two.
Embryos from one of the recovered mutants (88.21) lack isl1 expression in the endocrine pancreas, but maintain expression in the CNS ( Figure 1B versus wild type in Figure 1A). 88.21 embryos first display a morphological phenotype approximately at day 4, as they do not develop fin buds and have a swollen pericardium ( Figure 1D versus wildype in Figure 1C). Since the EK and TL strains used in our screen are highly polymorphic with respect to their CA repeats, we used a PCR panel consisting of 192 primer pairs that amplify CA repeats in the zebrafish genome to map the position of the mutation [36,37,38]. Specifically, genomic DNA pools from phenotypically wild type and phenotypically mutant embryos were amplified using primers from the CA marker panel. Based on the bulk segregant analysis of the DNA pools, two markers, z10441 and z8693, were found to be linked to the mutation ( Figure 1F). Subsequent PCR of inividual embryos (not shown) confirmed the linkage. We detected three crossovers out of 44 meioses for the z10441 marker, which places the mutation approximately 7 cM away from this marker on linkage group 7 ( Figure 1G).

The 88.21 Mutant Represents a Novel aldh1a2 Allele
A closer examination revealed that the z10441 and z8693 markers are both located near the aldh1a2 (raldh2) gene on chromosome 7. As noted, aldh1a2 is a retinaldehyde dehydrogenase (Raldh) involved in RA synthesis and there are two previously reported aldh1a2 mutants, neckless (nls or aldh1a2 i26 ; Figure 1E) and no fin (nof or aldh1a2 u11 ) [28,34]. Since the 88.21 mutant phenotype bears some resemblance to the aldh1a2 i26 phenotype ( Figure  1D, E) -lack of pectoral fins, swollen pericardium and embryonic lethality by day 6 -we tested if 88.21 might represent a novel aldh1a2 allele. To this end, we amplified full length aldh1a2 from cDNA prepared from mutant and wild type embryo pools derived from an 88.21 incross. Sequencing of the PCR products identified a G to A change in the mutant pool that converts a glycine to an arginine at position 484 ( Figure 1H, J) in the catalytic domain of Aldh1a2 ( Figure 1K). Sequencing cDNA from individual embryos confirmed this change ( Figure 1I).
To confirm that the 88.21 phenotype is caused by a mutation in the aldh1a2 gene, we set out to rescue the mutant phenotype with wild type aldh1a2 mRNA. We find that 26% of embryos from an incross of 88.21 heterozygotes fail to develop fin buds (Table 1), as assayed by shh expression in fin buds at 48 hpf (Figure 2A, B) or by visual inspection for fin bud formation at 72 hpf (not shown). However, following injection of wild type aldh1a2 mRNA at the 1-2 cell stage, only 8.5% of embryos lack fin buds, demonstrating that aldh1a2 mRNA rescues fin bud development (Table 1, Fig. 2C). In contrast, injection of aldh1a2 mRNA containing the 88.21 mutation does not rescue fin bud development (24% lack fin buds; Table 1, Fig. 2D). Notably, the swollen pericardium phenotype was not rescued by injection of aldh1a2 mRNA. This result is consistent with previous work showing that fin bud development in aldh1a2 i26 and aldh1a2 u11 can be rescued by injecting wild type aldh1a2 mRNA, but the swollen pericardium cannot be rescued [28,34]. We conclude that the 88.21 mutation occurs in the aldh1a2 catalytic domain and we refer to it as aldh1a2 um22 . Since the mutant mRNA appears to be inactive even when overexpressed, the aldh1a2 um22 allele is likely to represent a null allele. In particular, replacing a small conserved glycine residue with a large arginine in the catalytic domain may affect the function or folding of the Aldh1a2 protein.
Endoderm Gene Expression Is Variably Affected in aldh1a2 um22 and aldh1a2 i26 Mutants We observe variable effects on endoderm gene expression in aldh1a2 um22 mutants and we therefore compared the aldh1a2 um22 phenotype to the aldh1a2 i26 phenotype. The aldh1a2 i26 allele was previously analyzed with some endodermal markers [15] but we have expanded the analysis further. We find that endocrinespecific genes such as isl1 (Table 2) and sst1 (Table 2) are completely lost in both mutants at 24-30 hpf, as is p48 expression in the exocrine pancreas (Table 2). In contrast, pdx1 expression remains in the majority of both aldh1a2 i26 and aldh1a2 um22 mutant embryos ( Figure 3E versus 3G, H; Figure 4A versus 4C, D; Table 2), as does carboxypeptidase A (carbA) expression, although carbA expression is more pronounced in aldh1a2 i26 ( Figure 4I versus 4K, L; Table 2). We also find that expression of hhex and prox1 (that are expressed in both the ventral pancreatic bud and the liver) persists in both mutants ( Figure 3I versus 3K, L and 3M versus 3O, P; Table 2). Analyzing other liver markers later in development revealed that expression of both sid4 (at 48 hpf) and transferrin (transf, at 72 hpf) persists in both aldh1a2 um22 and aldh1a2 i26 mutant embryos ( Figure 4E versus 4G, H; 4Q versus 4S, T; Table 2). intestinal fatty acid binding protein (ifabp) expression is decreased at 72 hpf ( Figure 4M versus 4O, P; Table 2) in aldh1a2 um22 and aldh1a2 i26 mutant embryos, suggesting that differentiation of the intestine takes place, although perhaps not to completion. Expression of the early endoderm marker sox17 is maintained (data not shown). Also, while our data suggest that endocrine gene expression may be most sensitive to the loss of Aldh1a2 function, we find that insulin (ins) expression remains in some mutant embryos at 24, 48 and 72 hpf, suggesting that endocrine gene expression is not completely blocked in the mutants (Fig. 3C, D; Table 2). Lastly, we tested whether embryos with residual expression of one endoderm gene had residual expression of other endoderm genes, but did not observe a correlation, suggesting that expression of each gene varies from embryo to embryo ( Figure S1).

The Aldh Inhibitor DEAB Completely Blocks Expression of Endoderm Genes
We reasoned that the residual gene expression observed in aldh1a2 um22 and aldh1a2 i26 mutant embryos could either indicate that RA signaling is not completely required for expression of all genes in the endoderm, or it might indicate residual Aldh activity in the mutants. To test this further, we made use of DEAB (diethylaminobenzaldehyde), a competitive reversible inhibitor of all Aldh enzymes. DEAB has previously been reported to block development of fin buds and otic vesicles [29] and blocks expression of hoxb1b, vhnf1, krx20 in rhombomere (r) 5, val in r5-6, hoxd4a and efnb2a in r7 of the hindbrain [42]. Zebrafish embryos treated with DEAB have been analyzed for a few endoderm markers [15,22,43,44,45]. In particular, insulin::GFP expression is lost in embryos treated with DEAB [15,43,44,45]. Also, foxa3 expression in the pancreas and liver and vhnf1 expression in the pancreas is lost in DEAB treated embryos [45]. Loss of pharyngeal arches 3-5 was also seen when DEAB was used [46]. We find that treating zebrafish embryos with 10 uM DEAB starting at 8 hpf (see Figure S2 for DEAB titrations) blocks endoderm gene expression. Specifically, expression of ins, pdx1, hhex, prox1, sid4, carbA, ifabp and transf is completely lost in DEAB treated embryos ( Figure 3B, F, J, N; Figure 4B, F, J, N, R; Table 2) while sox17 expression is unaffected (not shown). Notably, treatment with lower concentrations (1-5 uM) of DEAB closely mimics the phenotypes observed in aldh1a2 um22 and aldh1a2 i26 mutant embryos ( Figure S2). We conclude that Aldh activity is absolutely required for endoderm gene expression and that there is residual Aldh activity in aldh1a2 um22 and aldh1a2 i26 mutant embryos.
Maternal aldh1a2 Activity Persists in aldh1a2 um22 and aldh1a2 i26 Mutant Embryos We next considered the likeliest source of residual Aldh activity in aldh1a2 um22 and aldh1a2 i26 mutant embryos. The expression patterns of aldh1a3 (raldh3; observed primarily in developing eye, inner ear, pituitary gland and swim bladder) and aldh8a1 (raldh4; found in liver and intestine, but not until day 2) [26,27] make them unlikely candidates for providing Aldh activity in early pancreas development. In addition, raldh1 is expressed in the dorsal retina and mesencephalic flexure in mice [26], but has not been found in zebrafish. Instead, we reasoned that there may be residual aldh1a2 activity in the mutants. Since the aldh1a2 um22 and aldh1a2 i26 mutations are likely to be null mutations, we considered the most likely source of residual aldh1a2 activity to be maternally deposited mRNA.
To test this possibility, we first carried out RT-PCR on 3 hpf (before the onset of zygotic transcription) and 6 hpf (after the onset of zygotic transcription) zebrafish embryos. We find that aldh1a2 mRNA is present already at 3 hpf ( Figure 5A), consistent with a role for maternal aldh1a2 mRNA. We reasoned that if the residual aldh1a2 activity observed in the mutants is due to maternal mRNA, then blocking aldh1a2 translation with antisense morpholino oligonucleotides (aldh1a2 tMO) should produce the same phenotype as DEAB treatment. Indeed, we find that injecting aldh1a2 tMO, completely blocks expression of hhex ( Figure 5C), prox1 ( Figure 5E) and pdx1 (Figure 5G), producing a phenotype indistinguishable from the DEAB phenotype and more severe than the aldh1a2 mutant phenotype, while embryos injected with a mismatch MO control show wild type expression of all endoderm markers ( Figure 5B, D, F). In contrast, we find that a MO targeting the aldh1a2 exon 1/intron 1 splice junction (which should not affect already spliced maternal aldh1a2 mRNAs) cannot fully block endoderm gene expression even at the highest concentration that could be tested (750 uM, not shown). We conclude that aldh1a2 is the predominant aldh required for RA signaling during endoderm development and that aldh1a2 has a significant maternal component.

Discussion
We report results from an ENU (N-ethyl-N-nitrosourea) screen for genes involved in endocrine pancreas development. We characterize the aldh1a2 um22 allele, which corresponds to a glycine to arginine mutation in the catalytic domain of the Aldh1a2 protein. aldh1a2 um22 mutant embryos show similarities to embryos of two previously identified aldh1a2 mutants, neckless (nls or aldh1a2 i26/i26 ) and no fin (nof, aldh1a2 u11/u11 ) [28,34] in that all three mutants do not develop fin buds and have a swollen pericardium. We compare the endoderm phenotype of aldh1a2 i26 and aldh1a2 um22 mutant embryos to that of embryos treated with DEAB (a pan-Aldh inhibitor). Interestingly, endoderm markers are not uniformly lost in aldh1a2 mutant embryos, but are lost in DEAB-treated embryos, suggesting residual Aldh activity in the mutants. We detect the presence of maternal aldh1a2 transcripts and demonstrate that a morpholino targeting the aldh1a2 translation start site copies the DEAB phenotype. We conclude that Aldh1a2 is the predominant Aldh enzyme acting in early pancreas development and that there is a significant role for maternally derived Aldh in this process.

Aldh Activity Is Required for Pancreas Development
Disrupted RA signaling has broad effects such as shorter body length, curved body axis, lighter pigmentation, immobility, and a swollen pericardium. As a result, many developmental defects are observed, including neural crest cell death, the absence of limb buds and posterior branchial arches, small somites, and hindbrain segmentation defects, which have been known in general as VAD (vitamin A-deficiency syndrome) [47,48,49]. In mouse, a null mutation in the Aldh1a2 gene mimics the hindbrain phenotypes associated with full VAD, establishing Aldh1a2 as the main RA producing enzyme required in hindbrain development [29,30,50,51]. As a result to losing RA, rhombomeric and gene expression boundaries posterior to rhombomere (r) 3 are lost [46,47,52,53,54,55,56]. In zebrafish embryos that are treated with DEAB to block Aldh activity, defects in anterior-posterior patterning of the neural tube also resemble severe VAD cases. The neural tube is strongly anteriorized and hindbrain development posterior to r4 is stopped. Also, loss of fin buds and reduction of branchial arches are observed [28,34,57]. This indicates a conserved role for Aldh enzymes in the production of RA required for hindbrain development in both zebrafish and mice.
RA is also involved in endoderm development in vertebrates. In mice, Aldh1a2 is expressed in the dorsal pancreatic mesenchyme during pancreas specification and RA-responding cells reside in both pancreatic endoderm and mesenchyme [58]. As a result, defects in the endoderm are observed in the absence of RA. In particular, Aldh1a2 2/2 mice lack Pdx1 and Prox1 expression in the dorsal pancreatic bud but the ventral bud appears normal [2,58]. Accordingly, Insulin and Glucagon-expressing cells do not develop and Isl1 expression is severely decreased [58]. Hlxb9, expressed in the dorsal foregut endoderm, is also reduced [58]. Expression of Foxa2 in the dorsoventral axis of the endoderm is not affected, indicating that early endoderm development is unaltered [2]. Hhex expression is not affected in the liver, suggesting that RA is not involved in liver development -similar to observations in Xenopus and avian embryos [2,12,13]. Treating Xenopus embryos with a RA receptor antagonist (BMS493) blocks dorsal pancreatic development, but does not affect ventral pancreatic development or the liver [12]. Similarly, in RA-deficient avian embryos or VAD (obtained from birds fed on a retinoid-deficient defined diet [59]), dorsal pancreas is lost but not ventral pancreas or liver [7,48,59]. Since Xenopus embryos treated with BMS493, VAD quail embryos and Aldh1a2 2/2 mutant mice display a similar phenotype -loss of dorsal pancreas but not ventral pancreas or liver -it appears that Aldh1a2 is the only Aldh acting in endoderm and that it is only necessary for dorsal pancreas development in these species. In contrast, blocking RA completely in zebrafish embryos eliminates all pancreas and liver gene expression. Embryos treated with DEAB lose vhnf1 expression in the pancreas, insulin:GFP expression in the endocrine pancreas, foxa3 expression in the pancreas and liver, and pharyngeal arches 3-5 are lost as well [15,22,43,44,45]. We treated embryos with 10 uM DEAB at 8 hpf and found that various endoderm markers expressed in the pancreas, liver, and intestine are lost, similar to embryos treated with BMS493 (pan-RAR antagonist) [15]. Also, injecting aldh1a2 translational MO (tMO) knocks down insulin expression [8] and we find that aldh1a2 tMO knocks down expression of genes such as hhex (liver and pancreas), prox1 (liver and pancreas), and pdx1 (pancreas and duodenum) as well ( Figure 5).
Thus, there appears to be a conserved role for RA in pancreatic development among vertebrates, but mouse, Xenopus and avian embryos have restricted RA's role to the dorsal pancreas. The liver and ventral pancreas emerge adjacent to one another from the ventral endoderm in a default state as pancreas, but the liver receives signals from the cardiac mesoderm (FGF) to express liver markers [60]. Interestingly, the markers that continue to be expressed in aldh1a2 um22 and aldh1a2 i26 mutant zebrafish embryos are those expressed in the ventral pancreas and liver (hhex, prox1, sid4, carbA, and transf), indicating that less RA is needed to turn on expression of these genes, possibly consistent with an evolutionary phasing out of RA's involvement in these regions. Therefore, RA's role in ventral pancreas and liver development does not appear evolutionarily conserved among vertebrates. Other signaling factors may have taken precedence over RA in development of these regions in mouse. For instance, BMP and FGF signaling is necessary for liver development in mouse embryos, but inhibiting FGF and BMP signaling in zebrafish embryos leads to a decrease, not a loss, of hhex and prox1 expression [61,62,63,64].
Lastly, treatment with DEAB does not affect early endoderm gene expression in zebrafish embryos (sox17) or mutant mouse embryos (FoxA2) [2], indicating a conserved role that RA is not necessary for early endoderm development in vertebrates.
The aldh1a2 i26 , aldh1a2 u11 and aldh1a2 um22 Alleles Likely Represent Null Mutations The zebrafish aldh1a2 mutant alleles exhibit defects in patterning of the neural tube and the endoderm, although the phenotype is not as severe as in DEAB-treated zebrafish embryos (Figs. 3, 4) [28,34], Aldh1a2 2/2 mutant mice or VAD quail and rat embryos [28,34]. Instead, it is similar to the phenotype we observe upon treatment with a low concentration of DEAB ( Figure S2), as well as to a mild version of VAD seen in rat embryos and to partial rescue of Aldh1a2 2/2 mouse embryos by maternal application of RA [51,55,65]. Since Aldh activity appears absolutely required for pancreas formation (because DEAB-treated embryos lack endoderm gene expression, see above), the weaker phenotype of aldh1a2  Table 2 for quantification. doi:10.1371/journal.pone.0008261.g003 mutant zebrafish embryos could be explained if the aldh1a2 i26 , aldh1a2 u11 and aldh1a2 um22 alleles represent hypomorphic mutations that maintain some residual Aldh activity.
However, the mutations occurring in the aldh1a2 i26 , aldh1a2 u11 and aldh1a2 um22 alleles appear likely to be null mutations. In each case, the mutated residue is conserved across human, mouse, rat, Xenopus, and zebrafish [28,34], indicating that amino acid sequence is important for the overall function of the Aldh1a2 protein and changing it will most likely affect the protein function. Furthermore, in each case, the mutation introduces a large charged residue (Gly -. Arg in aldh1a2 i26 , Thr -. Lys in aldh1a2 u11 , Gly -. Arg in aldh1a2 um22 ). Such replacements are likely to affect the proper folding of the protein and therefore affect the catalytic function of Aldh1a2.
Further support for the idea that aldh1a2 i26 , aldh1a2 um22 and aldh1a2 u11 represent null mutations comes from rescue experiments, which indicate that the mutant proteins are not functional. When we injected aldh1a2 um22 embryos with mRNA containing the aldh1a2 um22 mutation, it could not rescue fin bud development ( Figure 2 and Table 1). However, when we injected wild type aldh1a2 mRNA, we were able to rescue fin bud development. The same was seen in rescue experiments using both aldh1a2 i26 and aldh1a2 u11 [28,34]. Furthermore, overexpression of the aldh1a2 um22 mutant mRNA in zebrafish embryos did not affect development (not shown), further demonstrating that the aldh1a2 um22 allele is inactive. Together, this indicates that the aldh1a2 i26 , aldh1a2 um22 and aldh1a2 u11 mutations do not result in hypomorphic proteins, but represent null mutations.

A Role for Maternal aldh1a2 mRNA
If the aldh1a2 i26 , aldh1a2 u11 and aldh1a2 um22 alleles encode inactive Aldh1a2, the fact that aldh1a2 mutant zebrafish embryos do not display a severe VAD phenotype suggest that Aldh activity must be coming from another source. The expression pattern of other aldhs rules them out as likely candidates and we therefore focused on maternal aldh1a2 mRNA. We find that aldh1a2 is expressed already at 3 hpf, albeit at somewhat lower levels -this lower level may explain the weak phenotype observed in the mutants. We also find that a MO targeting the exon 1/intron 1 splice site of aldh1a2 (sMO, which should target only zygotic transcripts) produces a milder phenotype (not shown) and that lower doses of aldh1a2 tMO (500 uM) permit some expression of  Table 2  pdx1, similar to the phenotype observed in aldh1a2 um22 and aldh1a2 i26 mutant embryos). We also note that after treating with DEAB, it became clear that aldh1a2 i26 embryos still have some RA activity since DEAB treated embryos display a severe phenotype similar to VAD [34,57,65]. This residual aldh1a2 is most likely due to maternally supplied mRNA.
In contrast, Aldh enzymes do not appear to be deposited maternally in other vertebrates. In particular, the fact that aldh1a2 mutations in mice mimic the VAD phenotype [47,48,49], suggests that there are no maternally contributed Aldhs in the mouse. In Xenopus, both retinol and retinaldehyde are present in embryos before gastrulation, indicating that RDHs may be present (possibly maternally deposited) [57,66]. Furthermore, microinjection of Aldh1 or Aldh1a induces premature RA signaling in Xenopus [67] by acting on this retinaldehyde pool, suggesting that maternally deposited Aldhs are not present in the Xenopus embryo [28,50,67,68]. It is not clear why aldh1a2 is maternally deposited in zebrafish, but our observation that treatment with DEAB before gastrulation results in death or severely deformed embryos (not shown) suggests that there may be an early role for aldh1a2 in zebrafish embryos.