Genomewide Expression Analysis in Zebrafish mind bomb Alleles with Pancreas Defects of Different Severity Identifies Putative Notch Responsive Genes

Background Notch signaling is an evolutionarily conserved developmental pathway. Zebrafish mind bomb (mib) mutants carry mutations on mib gene, which encodes a RING E3 ligase required for Notch activation via Delta/Jagged ubiquitylation and internalization. Methodology/Principal Findings We examined the mib mutants for defects in pancreas development using in situ hybridization and GFP expression analysis of pancreas-specific GFP lines, carried out the global gene expression profile analysis of three different mib mutant alleles and validated the microarray data using real-time PCR and fluorescent double in situ hybridization. Our study showed that the mib mutants have diminished exocrine pancreas and this defect was most severe in mibta52b followed by mibm132 and then mibtfi91, which is consistent with the compromised Notch activity found in corresponding mib mutant alleles. Global expression profile analysis of mib mutants showed that there is a significant difference in gene expression profile of wt and three mib mutant alleles. There are 91 differentially expressed genes that are common to all three mib alleles. Through detailed analysis of microarray data, we have identified several previously characterized genes and some putative Notch-responsive genes involved in pancreas development. Moreover, results from real-time PCR and fluorescent double in situ hybridization were largely consistent with microarray data. Conclusions/Significance This study provides, for the first time, a global gene expression profile in mib mutants generating useful genomic resources and providing an opportunity to identify the function of novel genes involved in Notch signaling and Notch-regulated developmental processes.


INTRODUCTION
The Notch pathway is an evolutionarily conserved signal transduction cascade that plays essential roles in a variety of developmental processes, such as pattern formation, cell fate determination and organ formation through local cell-cell interactions (reviewed in [1][2]). Apart from being important for normal development, Notch signaling is also related to several human congenital diseases, such as T-cell acute lymphoblastic leukemia/lymphoma [3][4][5], Alagille syndrome [6][7][8], a late onset neurological disease (CADASIL) [9], and spondylocostal dysostosis [10].
Microarray is a useful method for genomewide expression profile analysis [48]. With this approach, several downstream genes of signaling pathways involved in controlling mouse pancreas development have been reported [49,50]. Following the leads from such microarray analysis, a novel transcription factor, Myt1 [50] has been shown to be involved in pancreas development. Similarly, Mellitzer et al. also showed that a novel transcription factor, IA1, is Ngn3-regulated and required for normal differentiation of endocrine pancreatic cells [51]. Highdensity microarrays have been established for zebrafish and transcriptome profiles during embryogenesis have been documented [52]. In addition, new target genes involved in sonic hedgehog signaling pathway has been identified in zebrafish using microarrays [53]. However, a genomewide analysis of Notch signaling defective mutants has not been reported.
Hence, we have carried out this study with three major objectives, viz. (1) to examine the three mib mutant alleles for pancreas defects and understand the consequence of differential activation of Notch signaling on pancreas development using in situ hybridization and pancreas-specific GFP expression analysis, (2) to identify differentially regulated genes between the wt zebrafish and three different mib mutant alleles through microarray analysis and (3) to identify and validate the Notch-responsive genes involved in pancreas development. Through our present work, we have shown that the development of exocrine pancreas is diminished in mib mutants and this defect is most severe in mib ta52b followed by that in mib m132 and mib tfi91 . Using microarray analysis, we have identified several characterized genes, and novel genes/ESTs potentially involved in pancreas development. Validation of representative genes with real-time PCR and fluorescent double in situ hybridization supported the microarray data. Furthermore, this study has generated useful genomic resources identifying a number of uncharacterized ESTs/genes that may play a significant role in Notch-regulated developmental process.

Pancreas defects in mib mutant alleles
We carried out in situ hybridization for four pancreas-specific genes in mib mutants of 4-day post-fertilization (dpf) and their wild type (wt) siblings ( Figure 1). The expression of exocrine pancreasspecific genes, elastaseA ( Figure 1A and 1A9-1D9) and trypsin ( Figure 1E and 1E9-H9) were diminished in mib mutants. In comparison to their expression in wt embryos, these two genes were highly down-regulated in mib ta52 followed by those in mib m132 and then mib tfi91 , which are less down-regulated. This observation is in concurrence with the diminished Notch activity in these mib mutants [34]. We also analyzed the elastaseA-GFP expression in 4dpf mib mutants and their wt siblings (Figure 2A-2D). The elastaseA-GFP expression was diminished in mib mutant alleles in the same order of decreasing Notch activity in these mutants, which corroborates our observation shown by in situ hybridization ( Figure 1A and 1A9-1D9).
We further analyzed the expression of three endocrine pancreasspecific genes, somatostatin (d-cell-specific), insulin (b-cell-specific), and pdx1 (pancreas progenitor-specific). Levels of somatostatin ( Figure 1I and 1I9-1L9) and insulin ( Figure 1M and 1M9-1P9) were slightly increased in 4-dpf mib mutants. The insulin-GFP ( Figure 2E and 2E9-2H9) and pdx1-GFP ( Figure 2I and 2I9-2L9) expression analysis in mib mutants also showed that these genes are slightly up-regulated compared to their expression in wt embryos.

Genomewide expression profiling
We carried out the genomewide expression profiling of mib mutants at three different stages-24-, 48-and 72-hour post-fertilization (hpf). Through microarray analysis of 72-hpf embryos, we identified 1128 up-regulated and 936 down-regulated genes in mib ta52b mutants (Table S1, q = 0.0); 1464 up-regulated and 2210 down-regulated genes in mib m132 mutants (Table S2, q = 0.0); and 2081 up-regulated and 2538 down-regulated genes in mib tfi91 mutants (Table S3, q = 0.0). Using a PERL script, we further identified the differentially expressed genes specific to each mutant allele and common to all three mib mutant alleles ( Figure 3). In the list of up-regulated genes, the numbers of genes specific to mib ta52b , mib m132 and mib tfi91 mutant alleles were 93, 287 and 768, respectively; in the list of downregulated genes, the numbers of genes specific to mib ta52b , mib m132 and mib tfi91 alleles were 33, 557 and 874, respectively (Figure 3A and 3B,  Table S5, S6 and S7, q = 0.0 and score(d).4.0). The majority of these differentially expressed genes were uncharacterized genes or ESTs (Table 1). There were 91 genes common to all three mib  Table 1). Of these 91 genes, only 27 genes were previously characterized and 64 were uncharacterized genes or ESTs (Table 1, Table S4, q = 0.0 and score(d).4.0). The cluster tree view showed the expression profile of these 91 genes: the up-regulated ones were shown in red and the down-regulated ones in green ( Figure 4). We further categorized these 91 common genes according to their known function or predicted function based on their homology to mouse and human orthologs (Table S4). Of the 27 characterized genes, there are 6 up-regulated genes, including dab2, mcl1a, mcl1b, fn1l, ttn and nppa, and 21 down-regulated genes, including pbx3b, gpm6aa (BI840762), olig2, tfdp2, rpl13, gpm6aa (BI839927), atp1a1b, fabp7a, gpm6aa (BG306150), fkbp5, gfap, opn1sw1, opn1sw2, opn1mw1, vsx1, pou50, mdkb, tal1, dla, vamp2 and her4.
Using the same PERL script, we analyzed the 48-hpf time point data for three mib mutant alleles, and data of 24-hpf, 48-hpf and 72-hpf time points for the mib ta52b mutant allele. Numbers of differentially expressed genes common to all three mutant alleles or three time points, and specific to each mutant allele or time point are shown in the Venn diagram ( Figure 3C-3F). At 48 hpf, there are 44 (30 up-regulated and 14 down-regulated) differentially expressed genes common to all three mutant alleles, and 129 (54 upregulated and 75 down-regulated), 124 (44 up-regulated and 80 down-regulated) and 693 (635 up-regulated and 58 down-regulated) genes specific to mib ta52b , mib m132 and mib tfi91 mutant alleles, respectively (Figure 3C and 3D; Table S8, S9, S10 and S11, q = 0.0). For mib ta52b mutants, there are 61 (32 up-regulated and 29 down-regulated) genes common to all three time points, and 1821 (1128 up-regulated and 693 down-regulated), 49 (14 up-regulated and 35 down-regulated) and 1851 (930 up-regulated and 921 downregulated) genes specific at 24 hpf, 48 hpf and 72 hpf, respectively (Figure 3E and 3F; Table S12, S13, S14 and S15, q = 0.0).

Functional categories
We used the microarray data of three mib mutant alleles (q = 0.0, score(d).4.0) and the 91 common genes at 72 hpf for functional analysis. We first searched for the gene ontology and the functional similarity of their human and mouse orthologs in the Zebrafish Chip Annotation Database. Then, we classified them into different functional categories based on their known functions in zebrafish or in mouse and human orthologs, such as 'transcription factor/    Finally, we plotted them in a pie chart format either including the uncharacterized genes and ESTs ( Figure 5A-5D) or excluding them ( Figure 5E-5H). The majority (.70%) of the differentially expressed genes are uncharacterized genes or ESTs ( Figure 5A-5D). Among the genes with known or related functions in the mib ta52b data set, about 35% belong to the category of 'transcription factor/ nuclear' related genes and the remaining genes to other categories, such as 'transport proteins' (25%), 'signaling' (19%), 'structural proteins' (8.6%), 'cell adhesion/matrix' (5.4%), 'hormone activity' (4.3%) and 'cell cycle/apoptosis' (3.2%) ( Figure 5E).

Functional groups and pathway analysis by IPA
To further analyze our data set, we used IPA (see Materials and Methods) to identify functional groups and selected the top 15 ones as significantly enriched for their respective functions ( Figure 6A). The significance of each function (calculated from negative log of p-value; -log 1.3 is equal to p-0.05) revealed that all the top 15 functions are highly and significantly enriched in all three mib alleles. Some of the significantly enriched functions are 'cellular growth and proliferation', 'cellular development', 'gene expression' and 'embryonic development'. Genes related to 'connective tissue disorders', 'skeletal muscular disorders' and 'cancer' are significantly enriched in the mib ta52b allele.
One of the tools in IPA enables us to identify the enrichment of genes in selected canonical pathways from the input data set. The top 15 canonical pathways that are enriched in the mib mutants include Notch and Wnt signaling pathways ( Figure 6B). Genes related to Wnt signaling were significantly enriched in mib m132 and mib tfi91 mutants. Notch signaling genes were enriched to a lesser extent and their enrichment was not significant. Canonical pathways, such as 'actin cytoskeleton signaling', 'integrin signaling', 'calcium signaling' and 'TGF-beta signaling', were enriched in mib m132 and mib tfi91 .
Notch and Wnt/b-catenin signaling were shown to be affected in mib mutants [27,54,55]. We mapped the genes that are involved in these two signaling pathways and differentially expressed in different mib alleles. The genes involved in Notch signaling, such as dll1, notch1, notch2, hey1 and herp, were down-regulated in one, two or three mib alleles ( Figure S1A). In Wnt/b-catenin signaling, some genes, such as nlk, cx43, sox2, sox3, sox4, sox9, sox11, tcf7 and hdac1, were down-regulated in one or three mib alleles; while sox8 and tab1 were up-regulated ( Figure S1B). Few of the genes differed in the direction of expression change in different mutant alleles. For example, b-catenin was up-regulated in mib m132 while it was downregulated in mib ta52b and mib tfi91 ( Figure S1B).

Putative Notch signaling related genes
As positive controls of our microarray data, we searched for the known Notch signaling related genes in our list of differentially expressed genes. As expected, heyL was up-regulated and dla, her4, her8a, hes5, hey1, neurod, neurod4, notch1a, notch2 and notch3 were down-regulated (Table 2). To identify novel genes related to Notch signaling, we searched for the differentially expressed genes occurring at least two times in the gene sets common to all three mib mutant alleles at 72 hpf and 48 hpf, and the gene set common for the mib ta52b mutant allele at 72 hpf, 48 hpf and 24 hpf. We identified 31 such genes, the majority of which are novel genes or ESTs (Table S16). Of these 31 genes, only eight genes are characterized and five of them are known to be directly related to Notch signaling: olig2 [56], her4 [57], hes5 (also known as her15.1, [58,59]), dla [60] and nort ( [61], BI886648 is 98% identical to AB259590). Moreover, fn1l (also known as fn1b or fn3), in conjunction with its homolog, fn1, has been shown to function cooperatively with Notch signaling via integrina5 in somitogenesis [62,63]. gfap and gefiltin are glia-and neuron-specific structural proteins respectively and, therefore, indirectly linked to Notchmodulated gliogenesis and neurogenesis [64][65][66]. These observations suggest that a certain proportion of the 23 novel genes/ESTs are very likely to be involved in or related to Notch signaling.

Fluorescent double in situ hybridization validation of pancreas development related genes
To validate the microarray data for the genes predicted to be related to or involved in pancreas development, we carried out fluorescent double in situ hybridization at 3-dpf embryos for five down-regulated genes: trypsin, isl1, cad, wu:fb59c09 and notch1a, and five up-regulated genes: insulin, isl3, spon1b, glo1 and txnip (Table 3, Table 4 and Table  S17). Exocrine trypsin and endocrine insulin are predictably downand up-regulated, respectively ( Figure 7). However, notch1a, glo1 and txnip were not detected in the pancreas (see Discussion).
Down-regulated genes isl1 encodes a zebrafish insulin gene enhancer binding protein and it has 81% identity to mouse LIM/ homeodomain transcription factor ISL1 (islet-1). isl1 was expressed in nervous system, liver (data not shown) and pancreas of 3-dpf embryos. Our result is similar to that in earlier studies, which showed that isl1 is expressed in organs such as nervous system, liver and pancreas at various developmental stages of zebrafish [58,[72][73][74][75][76][77]. Its expression was strongly diminished in the nervous system (data not shown) of mib ta52b mutants; however, its expression was only slightly reduced in the endocrine pancreas compared to wt embryos ( Figure 7A and 7B). Our microarray results also showed that isl1 is down-regulated in mib ta52b mutants (Table 3).
cad encodes zebrafish carbamoyl-phosphate synthetase 2. Microarray analysis showed that cad is down-regulated (Table  S17). cad was expressed in endodermal organs, such as pancreas, intestine and liver, which is similar to that observed in earlier reports [78,79]. In mib ta52b mutants, cad expression was downregulated in all these tissues (Figure 7C and 7D; data not shown).
wu:fb59c09 is a zebrafish EST with significant similarity to a hypothetical gene, LOC570477. The human ortholog of this gene is Peroxiredoxin 4 (prx4). In situ hybridization results showed that this gene is expressed in liver, pancreas, intestine and other endodermal tissues (data not shown). It was down-regulated in both endocrine and exocrine pancreas of mib ta52b mutants compared to that in their wt siblings ( Figure 7E and 7F).
Up-regulated genes spon1b encodes an extracellular matrix protein, whose C-terminal contains five repeats identified previously in Thrombospondin and other proteins implicating in cell adhesion. Fluorescent double in situ hybridization results showed that this gene is expressed in endocrine pancreas, similar to earlier studies [77,79], and was up-regulated in mib ta52b mutants compared to that in their wt siblings ( Figure 7G and 7H).
isl3 encodes a zebrafish insulin gene enhancer binding protein homolog. It has high similarity to human ISL1 transcription factor, LIM/homeodomain (Islet-1) and ISL2 transcription factor, LIM/homeodomain (Islet-2). isl3 was expressed in endocrine pancreas, liver and intestine and its expression was up-regulated in mib ta52b mutants (Figure 7I and 7J; data not shown).

Real-time PCR validation of global expression profile
To further validate the microarray expression profile analysis, we carried out real-time PCR for 25 genes using the primers listed in Table S18. All the PCRs yielded the expected 300 base pair products. The real-time PCR ratios are basically similar to those of microarray data (Table 5), which confirms the validity of the array data.
We noticed that the results from two methods for few genes are not consistent. Therefore, to further address this question statistically, we evaluated the linear relationship between the microarray ratio (fold change) and the real-time PCR ratio for mib ta52b allele at 72 hpf using Pearson's correlation coefficient. It indicates a statistically significant positive correlation between these two sets of ratios (R = 0.914, R 2 = 0.835, p,0.001). This shows that the two sets of ratios are significantly (positively) correlated and the data are acceptably reliable [80].
DISCUSSION mind bomb mutants have compromised Notch activity due to mutations in the mib gene and serve as a unique resource to study the role of Notch signaling on various developmental processes [27,29]. Moreover, there are several mib alleles with different  [34]. In this study, we have examined the mib mutants for defects in pancreas development using in situ hybridization and pancreas-specific GFP expression analyses, and compared the global expression profile of three mib mutant alleles and their wt siblings using oligo microarray chips.
mib alleles are unique for studying Notch signaling The mib mutants have diminished exocrine pancreas development as evidenced by decrease in elastaseA and trypsin expression. The mib ta52b mutant allele showed maximal decrease in exocrine pancreas followed by mib m132 and mib tfi91 alleles. This indicates that there is a dose-dependent response of Notch signaling on pancreas development, since these alleles have different degree of compromised Notch activity [34]. Our GFP expression analysis in elastaseA-GFP lines also showed a similar effect of graded Notch signaling on pancreas development. Likewise, based on in situ hybridization results for exocrine pancreas-specific genes, such as mnr2a, ptf1a and trypsin in mib ta52b mutants and DAPT-treated mib ta52b mutants, Zecchin et al. concluded that a blockage of Notch signaling decreases the number of exocrine pancreatic cells [40]. Earlier study by Esni et al. also showed the role of Notch signaling in exocrine pancreas development of zebrafish [37].
The in situ hybridization results for the endocrine pancreasspecific genes, such as insulin and somatostatin, showed slight increase in their expression in mib mutant alleles; however, a dosedependent response of Notch signaling was not obvious (Figure 1). The GFP expression analyses in insulin-GFP and pdx1-GFP lines also showed a slight increase in endocrine pancreas in mib mutant alleles ( Figure 2). Similar to our results, Zecchin et al. also showed that in mib mutants (mib ta52b ) there is an increased expression of insulin and somatostatin [40].

Microarray analysis in mib alleles
In this study, we have used the microarray chips with 16,416 probes, representing 15,800 unique zebrafish genes. Earlier studies have successfully used this version of microarray chips for elucidating genes involved in zebrafish embryogenesis [52], liver tumor progression [81] and Hedgehog signaling [53]. Comparison of our microarray results for three mib mutant alleles at 72 hpf showed that the number of differentially expressed genes (q = 0.0, score(d).4.0) in mib tfi91 allele (2352) is greater than that in mib m132 (1536) and mib ta52b (321) alleles ( Figure 3A and 3B). Interestingly, this is inversely related to the severity of their phenotypes, such as fused somite borders, diminished tail pigmentation, neuronal hyperplasia and diminished her4 expression: most severe in mib ta52b allele followed by those in mib m132 and mib tfi91 alleles [34]. So far, there is no known molecular mechanism to explain this finding. Future studies need to be focused in this direction. We observed fewer differentially expressed genes at 48 hpf compared to that at 72 hpf for all three mib mutant alleles. This is in concurrence with the phenotypes of these three mib alleles, which are less obvious at 48 hpf compared to that at 72 hpf [34]. Owing to the large number of differentially expressed genes at these two time points, we narrowed our focus on the 72-hpf data set alone. In all three mib mutant alleles at 72 hpf, the majority (.70%) of the differentially expressed genes are uncharacterized genes or ESTs. This is because the zebrafish genome annotation has not been completed and the majority of genes remain as uncharacterized ESTs. Though a number of differentially expressed genes have been identified as full-length clones with ZGC IDs, they still remain unannotated. Gene Ontology (GO) analysis of differentially expressed genes of mib ta52b allele showed that less than 29% of the genes are characterized, the majority (35%) of which belong to the category of 'transcription factors/ nuclear' followed by the category of 'signaling molecules' (19%). This could be due to the fact that the mib mutants are defective in Notch signaling, which is one of the fundamental signaling pathways required for proper development of an organism.

Genes potentially involved in Notch signaling
Our microarray data is reliable, because some expected Notch signaling related genes did appear in the transcriptome, such as up-regulated heyL and down-regulated dla. Comparison of three sets of genes (72 hpf and 48 hpf for three mib alleles and 24 hpf, 48 hpf and 72 hpf of the mib ta52b allele) showed that there are 31 genes common to all three gene sets (Table S16). Five of these 31 genes, namely olig2 [56], her4 [57], hes5 (also known as her15.1, [58,59]), dla [60] and nort [61], are previously shown to be involved in Notch signaling. Particularly, based on similar global expression analyses of notch1a and notch3 morpholino morphants, Tsutsumi and Itoh showed that nort is a putative noncoding RNA regulated by Notch signaling in zebrafish [61]. Therefore, it is likely that the rest of the genes in this group are a part of the Notch connected network and hence serve as a useful resource to further identify novel genes working downstream of Notch.

Genes potentially involved in pancreas development
As a major focus of our current study, we searched for the genes related to pancreas development. As of now, there is no bioinformatic tool available to classify the zebrafish genes according to their biological functions. Therefore, we carried out a detailed manual search for the genes related to pancreas development. Using the ZFIN in situ hybridization expression database, we found 98 zebrafish genes that have been previously shown to be expressed in pancreas (Table 3, Table 4 and Table S17). The microarray expression profile for these genes showed that their expression profile is significantly different (q,2.0) in at least one of the three mib alleles. Several genes, such as neurod, isl1 and pax6b [82][83][84] were formerly shown to be responsive to Notch signaling in mice.
However, the fold change of expression for each gene is different in these three alleles. Only 25 out of these 98 zebrafish genes involved in pancreas development show significant difference (q,2.0) in expression profile in all three mib alleles. Out of these 25 genes, only 15 genes show consistent up-regulation (3 genes) or down-regulation (12 genes) in all three alleles. Apart from these 98 genes, we discovered 9 differentially expressed genes (Table S19), which are likely to play a role in zebrafish pancreas development based on their functional homology to mouse and human orthologs. It is definitely worth addressing in the future.

Is Mib Notch-specific?
So far, the experiments have unequivocally proved that Mib is an essential component of Notch signaling: it ubiquitylates and then endocytoses Delta with the extracellular part of Notch and therefore allows the intracellular part of Notch to enter the nucleus in the receiving cell and activate downstream target genes [27]. There is also evidence to suggest that Mib may be linked to Wnt signaling. Riley, et al. showed that heat shock-driven wnt1 expression in mib mutants leads to a partial rescue of its hindbrain metameric patterning phenotype [54]. Furthermore, knockdown of wnt3a and wnt8b in Dfw5 mutants, where wnt1 and wnt10b are deleted, resulted in the lost of boundary cells in hindbrain, which is similar to that in mib ta52b mutants [54]. However, this is in sharp contrast to wnt1 and tcf3b morphants, where the boundary cells are increased [55]. Therefore, the mechanism remains to be determined. From our IPA pathway analysis in all three mib alleles, we found enrichment of several differentially expressed genes belonging to various canonical pathways that have not been shown to directly link to Mib, including IGF-1, PDGF, EGF and IL-2 signaling pathways ( Figure 6B). This raises the possibility that Mib may be somehow involved in these signaling pathways. Furthermore, mouse DAPK and zebrafish Jagged2a have been shown to be substrates of Mib E3 ligase [85,86]. With a yeast two-hybrid screen, we also found that Mib binds to proteins involved in endocytosis and the ubiquitin-proteasome pathway (Chengjin Zhang, Jason Kin Wai Koo, Qing Li, Haoying Xu and Y.-J. J., unpublished data). Similarly, Snx5 has been identified as a Mibbinding protein from a yeast two-hybrid screen, which is colocalized with Mib in early endosomes and required for hematopoiesis and vasculogenesis [87]. All these observations suggest that Mib may not be Notch-specific and can also work with other pathways, just as previously shown for another E3 ligase, Itch, which targets Notch receptor and links to TNF through JNK [88][89][90]. Alternatively, the gene expression change could simply reflect the tissue/organ defects that are inflicted by a failure in Notch signaling. Our pathway analysis on microarray data supplies a good resource for examining whether Mib functions beyond Notch signaling and/or for testing what genes are involved in the tissue/organ mis-patterning caused by compromised Notch activity.

Validation of microarray data
To validate the microarray expression profile, we carried out fluorescent double in situ hybridization for ten genes (5 downregulated and 5 up-regulated) and real-time PCR for 25 genes. Out of these ten genes, insulin, trypsin, spon1b, cad, isl1, wu:fb59c09, notch1a, txnip and glo1 are previously shown to be involved in the development of and/or expressed in zebrafish pancreas; and isl3 is predicted to be related to pancreas development.
The microarray expression profile of trypsin and isl1 showed a down-regulation in mib ta52b allele and this was validated by our fluorescent double in situ hybridization and real-time PCR (Figure 7 and Table 5). wu:fb59c09 was slightly up-regulated (1.24) in microarrays of mib ta52b allele but it was down-regulated in mib m132 and mib tfi91 alleles (Table 3). However, our fluorescent double in situ hybridization and real-time PCR validations showed that it is down-regulated in the mib ta52b allele (Figure 7 and Table 5). cad expression was decreased in the mib ta52b allele by double in situ, which is consistent with the microarray data ( Figure 7 and Table S17). Earlier studies have shown that all these four genes are expressed in zebrafish pancreas at various developmental stages [37,75,78,79]. However, the role of Notch signaling on their expression profile in pancreas is not known except isl1. isl1 was shown to be up-regulated in primary neurons of mib mutants at 16s to 20s stage [91], but the effect of Notch signaling on its expression at later stages has not been studied. Furthermore, no observation has been made on its expression in pancreas. In contrast, our in situ hybridization results in 3-dpf embryos showed that the expression of isl1 in mib ta52b allele is down-regulated in nervous system but its expression in pancreas is only slightly reduced compared to that in wt embryos. In support of this observation, our microarray and realtime PCR results also showed that isl1 is down-regulated in mib ta52b mutants (Table 3 and 5).
Two up-regulated genes, insulin and spon1b, were validated. The microarray data showed that these two genes are up-regulated in the mib ta52b allele and this is supported by our fluorescent double in situ hybridization. isl3 was slightly increased by fluorescent double in situ hybridization. However, it was consistently decreased in mib ta52b allele by microarray analysis and real-time PCR (Figure 7, Table 4 and 5).
In contrast to the available information from ZFIN, which shows the gene expression pattern mainly up to 2 dpf, we did not detect notch1a, glo1 and txnip expression in the pancreas using fluorescent double in situ hybridization at 3 dpf, though we did detect notch1a expressed in, for example, hindbrain. It could be due to the stage difference or technical reasons. However, our real-time PCR results validated microarray expression for notch1a and glo1 ( Table 5).
The ratio comparison between microarray and real-time PCR for genes, such as ipf1, perp, igf2, isl3, spon1b and BI846588, showed that there is slight up-regulation in microarray, but downregulation in real-time PCR. However, it is evident that the overall expression alteration of all 25 genes is statistically comparable, although the actual fold change values in real-time PCR and microarray are not always commensurable. Such variation in values is likely due to the difference in sensitivity [80]. Nevertheless, our in situ hybridization results are highly consistent with our microarray profile.
In conclusion, the microarray analyses carried out in this study provide a useful resource of global gene expression profile of mib mutants defective in Notch signaling. Functional analysis of differentially expressed genes will shed light on their role in Notch signaling and various developmental processes.

MATERIALS AND METHODS Zebrafish wild type embryos and mind bomb mutants
We used AB strain wild type (wt) and three different mib alleles of different genetic severity, viz., mib ta52b , and mib m132 , mib tfi91 [34]. mib ta52b carries a missense mutation (M1013R) in the C-terminalmost RING finger domain; mib m132 carries a nonsense mutation (C785stop) leading to a truncated protein and mib tfi91 contains a nonsense mutation (Y60stop) [27]. mib ta52b and mib m132 are strong and weak antimorphic alleles, respectively, whereas mib tfi91 is a null allele [34]. All animal procedures were approved by the Biological Resource Centre, A*STAR.

Fish maintenance and sample collection
Fish were maintained in the IMCB zebrafish facility according to standard procedures. Crosses were set up in the evening and the barrier was lifted in the next morning. After half an hour, the fertilized embryos were collected and maintained at 28.5uC in egg water supplemented with methylene blue. For microarray analyses, the wt embryos were collected at 24 hpf, 48 hpf and 72 hpf, snap-frozen in liquid nitrogen and stored at -80uC. Mutants were separated from their wt siblings and frozen stored in the same way. At least two independent biological replicates were taken for each sample. All the samples were collected from the same cohort of fish to maintain a uniform genetic background.
For whole mount in situ hybridization and GFP analysis, the fertilized embryos were collected and grown at 28.5uC. Embryos were transferred to egg water with 0.033% phenylthiourea (PTU), The real-time PCR ratio in mib ta52b allele at 72 hpf is compared with the microarray data expressed as fold change values. The Genbank ID, Unigene name and Gene symbol were obtained using the Zebrafish Chip Annotation Database, (Unigene Build 85), http://giscompute.gis.a-star.edu.sg/,govind/zebrafish/version2/. which inhibits pigmentation, after 12 hpf and fixed/collected at appropriate stages.

In situ hybridization and fluorescent double in situ hybridization
DNA clones for making in situ hybridization probes were obtained from the Expressed Sequence Tag (EST) clone collection at the Genome Institute of Singapore (GIS) and the Institute of Molecular and Cell Biology (IMCB). Whole mount in situ hybridization using digoxigenin (DIG) (Roche) labeled RNA probes was carried out as previously reported [92]. Goat anti-DIG antibody conjugated to alkaline phosphatase (AP) was used for probe detection and NBT-BCIP was used as the substrate for color development. In situ hybridized embryos were observed using light microscope and the photos were taken with Zeiss Imager M1 microscope.
DIG-and fluorescein-labeled probes were generated via standard protocols. Embryos were proceeded for fluorescent double in situ hybridization with protocols previously reported [32,86]. However, the incubation temperature for probes was changed to 60uC to reduce the background. Photos were taken with Olympus Fluoview FV1000 microscope.

GFP expression analysis
To screen for defects in pancreas development, we have used three pancreas-specific GFP transgenic lines, viz., elastaseA-GFP (elaA-GFP) [93], insulin-GFP and pdx1-GFP lines [94]. These GFP lines were crossed with the heterozygous mib mutants and the offspring (F1) were grown until maturity. The F1 siblings carrying mib mutations were intercrossed to obtain embryos for GFP expression analysis. Photos were taken with Leica MZ FLIII microscope.

RNA extraction
Total RNA from the frozen embryos was extracted with Trizol (Gibco BRL) and cleaned with the Qiagen RNeasy mini kit. RNA quality was determined by gel electrophoresis, and the concentration was measured with a UV spectrophotometer. To reduce the bias, we used a common reference RNA for each time point, which was prepared at one time by extracting RNA from stagematched wild type embryos. The RNA extracts were stored at 280uC.

Microarray construction, target preparation and hybridization
The zebrafish microarrays were printed at GIS [52]. Oligonucleotide probes for this array were designed by Compugen (USA) and synthesized by Sigma Genesis (USA). For each gene, one 65-meroligonucleotide probe was designed from the 39 region sequence. Each probe was selected from a sequence segment that is common to a maximum number of splice variants predicted for each gene. The arrays contained 16,416 probes, representing 15, 800 unique zebrafish genes (UniGene build 85). In addition, the arrays also contained 170 spots representing b-actin gene probes as controls. The probes were suspended at a concentration of 20 mM in 3X SSC and spotted onto poly-L-Lysine coated microscope slides using custom-built DNA microarrayer.
For fluorescent labeling of target cDNAs, 20 mg of total RNA (10 mg, when the RNA quantity was limited) from reference and experimental samples were reverse-transcribed in the presence of Cy3-dUTP and Cy5-dUTP (Amersham Biosciences), respectively. Labeled target cDNAs were combined, concentrated and resuspended in DIG EasyHyb (Roche). Hybridizations on microarray slides were performed at 42uC for 16 h using MAUI Mixer FL (BioMicro Systems) as explained earlier [52,95]. At least two independent biological replicates were taken for each sample. At least two to three independent replicate hybridizations (technical repeats) were performed for each biological repeat sample (Table S20).

Scanning, filtering and data normalization
The arrays were scanned by the GenePix 4000B microarray scanner (Axon Instruments) to generate 16-bit TIFF image files. GenePix Pro 4.0 image analysis software (Axon Instruments) was used to measure the fluorescent signal intensity of the array features and local background on TIFF images. Only the gene features with signal background ratio more than 1.5 were used for analysis. The 16-bit TIFF image files and the gpr files with Cy3 and Cy5 signal intensities were uploaded into the GIS-developed Microarray Database (mAdb). Median normalization of the sample and reference channel intensity values was performed using the intensity-based log ratio median method [96]. The extracted intensity data from the mAdb database were normalized by Lowess normalization method and analyzed by modified tstatistic Significance Analysis of Microarrays (SAM) [97]. The microarray data files have been submitted to the Gene Expression Omnibus (GEO) and the accession number is GSE8522.

Gene annotation
The gene annotations were carried out by using the Zebrafish Chip Annotation Database (MySqL) (UniGene build 85) (http:// giscompute.gis.a-star.edu.sg/,govind/zebrafish/version2) developed and maintained by GIS. This database contains putative annotations for the probes in the zebrafish oligonucleotide array. We queried this database with the Genbank ID to obtain the following information: (1) Compugen description, (2) Zebrafish UniGene ID (build 85), (3) Zebrafish UniGene description (build 85), (4) Entrez Gene description, (5) Entrez Gene ID and Gene symbol, (6) GO term, (7) Locus Link, (8) UniGene protein similarity and description (mouse and human), (9) Full-length or assembled sequences, (10) HomoloGene (human and mouse, build 38.1) and (11) chromosomal location of the gene. Full-length sequences or the longest available gene sequences were obtained from the NCBI UniGene database (http://www.ncbi.nlm.nih. gov/sites/entrez?db = unigene&cmd). The Genbank IDs are referred to as 'genes' throughout this article.

Microarray data analysis
We applied Significance Analysis of Microarrays (SAM) [97] to identify statistically significant genes in each case. Since we have different number of replicates for different alleles, we used different thresholds (q = 0.0, score(d).4.0 [this represents that the absolute value of score(d) is greater than 4.0, namely, |score(d)|. 4.0] at 72 hpf; q = 0.0 at 48 hpf; q = 0.0 at 24 hpf) to select similar number of genes. However, the thresholds we used are all stringent (in every case the q-value is less than 2) and hence the false discovery rate (FDR, value expressed in %) in each case does not exceed 5. The FDR indicates the outcome with which the gene selected to be differentially expressed by the SAM analysis is likely to be occurring by chance. The score(d) indicates a statistic parameter, which is numerator(r) divided by denominator(s+s0) and hence serves as a cut-off point along with the q value. The numerator(r) value indicates the actual gene expression change shown as log2 value.
For the analysis of 72-hpf time point data, we selected those genes with q = 0.0 and scored(d).4.0 from the SAM generated data for all three mib alleles (Table S1, S2 and S3). The description for all the gene sets was obtained from the Zebrafish Chip Annotation Database [52]. Based on the most recently available information of the zebrafish gene annotation, these gene sets were classified as characterized and uncharacterized genes (Table 1). Furthermore, we manually searched for two sets of genes: one with functions related to Notch signaling and the other with functions related to pancreas development (Table 2-4). The expression profile values (log2) for the genes involved in the Notch pathway and pancreas development were obtained from the SAM analyzed data set. The in situ hybridization gene expression data for the genes related to pancreas development were obtained from the ZFIN database (http://zfin.org/cgi-bin/webdriver?MIval = aa-xpatselect. apg). If the gene was previously shown to be involved in pancreas development, we classified them as the 'genes involved in pancreas development'. If there is no functional relationship to zebrafish pancreas development but only functional homology to the human or mouse genes related to pancreas development, we classified them as the 'genes predicted to be involved in pancreas development'.
A PERL script was used to identify the differentially expressed (q = 0.0, score(d).4.0) genes (Genbank IDs) that are common among all three mib alleles (Table S4), between two different alleles and specific to each allele (Table S5, S6 and S7) at 72 hpf, and to remove duplicate genes (Genbank IDs), if any. Using this method, a group of 91 genes common to all three mib alleles were classified based on their function and characterization status (Table S4). These 91 genes were hierarchically clustered with TreeView_-vers_1_60 software [98] and tree view image was generated using Adobe Illustrator (Fig. 4).
For the analysis of 48-hpf time point data, the gene sets with q = 0.0 were used. The same PERL script was also used here to find the common genes and the allele-specific genes (Table S8, S9, S10 and S11). The SAM data (q = 0.0) for the mib ta52b on all three time points (24 hpf, 48 hpf and 72 hpf) were analyzed and queried using the PERL script to find out the gene sets that are common to all time points and specific to each time point (Table S12, S13, S14 and S15).

Functional groups and pathway analysis
Differentially expressed genes of three mib mutant alleles were subjected to Ingenuity Pathways Analysis (IPA) to identify the enrichment of genes in specific functional groups and pathways (IPA, Version 4, IngenuityH Systems, http://www.ingenuity.com). The IPA accepts human UniGene IDs as one of the identifiers for data upload and analysis. For this reason, the differentially expressed genes of mib mutants were mapped to their human homologs using the HomoloGene database and zebrafish UniGene mapping tool established at the GIS (http://giscompute.gis.a-star.edu.sg/,govind/unigene_db/). Human homologs of up-and down-regulated genes of the mutants were analyzed by using IPA tools and the enrichment of functional categories and canonical pathways with reference to the Ingenuity Pathways Knowledge Base (IPKB) were documented. Initially, differentially expressed genes of the three mutant alleles were individually analyzed. Subsequently, the enrichment patterns were compared among the mutants to identify the conservation of functional groups among the mutants.
Using the input data set (human homologs of zebrafish genes differentially expressed in the mib mutants), IPA identified a set of genes that are enriched for a particular function or pathway and the enrichment is represented as ratio. The ratio refers to the number of input genes associated with each function/pathway versus the total number of genes (available in IPKB) involved in that particular function/pathway. The ratios may be affected by the variations in the total number of input identifiers. In order to find the significance of enrichment in a particular function, IPA calculates the significance value based on the measure of involvement of the gene in the input data set to their respective molecular functions/signaling pathways. Using the right-tailed Fisher's Exact Test, the p-value (significance) is calculated by comparing the number of user-specified genes of interest that participate in a given function or pathway, relative to the total number of occurrences of these genes in all functional/pathway annotations in the IPKB.

Real-time PCR
To validate the microarray results, we carried out the real-time PCR for 25 genes and beta-actin gene was used as a reference. The primers used for amplifying each gene were listed in Table S18. cDNA was generated using the same purified RNA preparations from 72-hpf embryos (one biological repeat of wt and two biological repeats of mib ta52b ) used in microarray, and two other biological repeats of reference RNA from 72 hpf and one biological repeat sample RNA from mib ta52b mutants. RT-PCR was carried out using the LightCyclerH FastStart DNA Master-PLUS SYBR Green kit (Roche) and the Light Cycler machine as per the instructions of the manufacturer. The products of the RT-PCR were analyzed on the agarose gel electrophoresis for a single band of expected size. Relative cDNA amounts were calculated using the comparative C T method as explained in the real-time PCR manual of Applied Biosystems and normalized to the expression of beta-actin.

Analysis of correlation between microarray data and real-time PCR results
Subsequent to microarray and real-time PCR data analysis, an evaluation of linear correlation was performed for a set of 25 genes (Table 5), and the statistical significance of the correlation was determined using One-way ANOVA in SPS software. For the correlation analysis, the data input of the microarray was the fold change of expression and the data input of the real-time PCR was the ratio of relative expression for each gene. Both sets of ratios were obtained from the 72-hpf time point for the mib ta52b allele.