A systematic screen for genes expressed in definitive endoderm by Serial Analysis of Gene Expression (SAGE)

Background The embryonic definitive endoderm (DE) gives rise to organs of the gastrointestinal and respiratory tract including the liver, pancreas and epithelia of the lung and colon. Understanding how DE progenitor cells generate these tissues is critical to understanding the cause of visceral organ disorders and cancers, and will ultimately lead to novel therapies including tissue and organ regeneration. However, investigation into the molecular mechanisms of DE differentiation has been hindered by the lack of early DE-specific markers. Results We describe the identification of novel as well as known genes that are expressed in DE using Serial Analysis of Gene Expression (SAGE). We generated and analyzed three longSAGE libraries from early DE of murine embryos: early whole definitive endoderm (0–6 somite stage), foregut (8–12 somite stage), and hindgut (8–12 somite stage). A list of candidate genes enriched for expression in endoderm was compiled through comparisons within these three endoderm libraries and against 133 mouse longSAGE libraries generated by the Mouse Atlas of Gene Expression Project encompassing multiple embryonic tissues and stages. Using whole mount in situ hybridization, we confirmed that 22/32 (69%) genes showed previously uncharacterized expression in the DE. Importantly, two genes identified, Pyy and 5730521E12Rik, showed exclusive DE expression at early stages of endoderm patterning. Conclusion The high efficiency of this endoderm screen indicates that our approach can be successfully used to analyze and validate the vast amount of data obtained by the Mouse Atlas of Gene Expression Project. Importantly, these novel early endoderm-expressing genes will be valuable for further investigation into the molecular mechanisms that regulate endoderm development.


Background
The definitive endoderm (DE) is a population of multipotent stem cells allocated as one of the primary germ lay-ers during gastrulation. Initially formed as an epithelial sheet of approximately 500-1000 cells around the distal cup of an E7.5 mouse embryo, the DE is rapidly organized into a tube that runs along the anterior-posterior axis of the embryo [1][2][3]. The DE gives rise to the major cell types of many internal organs, including the thyroid, thymus, lung, stomach, liver, pancreas, intestine and bladder. Most of these organs have secretory and/or absorptive functions and play important roles in controlling body metabolism. Interest in the endoderm has intensified recently because processes that govern early development of DE-derived tissues may be recapitulated during stem cell differentiation [4,5], which could provide future therapies for diseased adult organs. Understanding how DEderived organs are specified, differentiate, proliferate, and undergo morphogenesis is key to understanding visceral organ disorders and tissue regeneration.
The last decade has yielded great insights into the molecular regulation of DE development [6]. In particular, pathways governing the initial formation of DE, patterning of the foregut, and morphogenesis of foregut-derived organs such as the pancreas and liver, have begun to be deciphered. Many of the key genes involved in the initial formation of DE are evolutionarily conserved. They include Nodal and components of its signaling pathway, transcription factors of the mix-like paired homeodomain class, Forkhead domain factors, and Sox17 HMG domain proteins [7][8][9][10][11]. Studies of ventral foregut patterning suggest that endoderm patterning is controlled by soluble factors provided by an adjacent germ layer [12]. FGF4, which is expressed in the neighboring cardiac mesoderm, can induce the differentiation of ventral foregut endoderm in a concentration-dependent manner [13,14]. FGF2 and Activin, secreted by the notochord, lead to the expression of pancreatic markers by repressing expression of Shh in pancreatic endoderm [15][16][17][18][19]. However, the precise hierarchical relationships between these factors and their downstream targets are still largely unknown, and complete molecular hierarchies have not been obtained. In addition, midgut and hindgut development is largely unexplored.
Embryonic stem (ES) cells have attracted much attention as a possible source of cells for regenerative medicine. Directing differentiation efficiently into specific lineages at high purities from ES cells requires both optimal selective culture conditions and markers to guide and monitor the differentiation process. While several methods of differentiation of ES cells to hepatic and insulin-producing cells have been described, determining the precise identity of these cells is problematic due to a lack of suitable markers [20][21][22][23]. More recently, two groups achieved efficient differentiation of human and murine ES cells into DE by combining directive culture conditions (serum concentration reduction and Activin supplements) and FACS sorting using the cell surface marker, CXCR4 [4,5,24]. Although useful, CXCR4 is not an ideal marker for the DE as it is widely expressed in the gastulation stage mouse embryo (Table 1 and [5,25]). At present there is no DEspecific marker that can unequivocally identify this cell type.
In summary, one major hurdle in the analysis of early DE development in both the embryo and ES cells is the lack of both pan-endodermal and endodermal region-specific genetic markers, since the majority of DE markers are also expressed in the visceral endoderm and/or other germ layers. Devising screens to identify genes specifically expressed in DE will contribute to studies of DE development. Several groups have carried out screens for novel genes expressed in the endoderm of Xenopus and mouse embryos using microarray or cDNA hybridization [25][26][27][28][29]. Despite the identification of several endoderm enriched genes, no novel DE specific genes were identified. As an alternative approach, we used Serial Analysis of Gene Expression (SAGE) to provide quantitative gene expression profiles. SAGE has been improved by the development of a longSAGE protocol, which generates tags that are 21 bp long and provides enhanced efficiency and accuracy of tag-to-gene mapping [30][31][32]. Compared with microarrays, SAGE has the additional advantage that it permits the identification of novel transcripts. SAGE also has the added benefit that the data are digital and thus can be easily shared among investigators and compared across different experiments and tissues.
In this study, we generated and analyzed three mouse DE longSAGE libraries. A list of candidate genes enriched for expression in endoderm was compiled through comparisons within these three endoderm libraries and against 133 mouse longSAGE libraries representing multiple embryonic stages and tissues generated by the Mouse Atlas of Gene Expression Project [32,33]. Sixty nine percent of these candidate genes showed previously uncharacterized expression in restricted tissues, including DE, after further whole mount in situ hybridization validation. Importantly, two genes identified, Pyy and 5730521E12Rik, showed exclusive DE expression at early stages of endoderm patterning. The high efficiency of this screen suggests that our endoderm libraries and the SAGE library database are powerful resources to identify tissue specific genes. Furthermore, these new endoderm genes provide a valuable tool for further investigation into the molecular mechanisms regulating endoderm development.

Overview of the endoderm libraries
Enriched definitive endoderm tissue was obtained by a combination of proteolysis and manual micro-dissection methods [14]. After removing the extra-embryonic region and digestion with trypsin, the DE was separated from ectoderm and mesoderm ( Figure 1A). Somite 0-6 endoderm pieces were pooled for the early whole endoderm library (SM108, Figure 1A). At this stage the newly formed endoderm has not yet been patterned, based on endoderm explant experiments [14,34]. Somite 8-12 endoderm was divided through the midgut into foregut and hindgut regions, and then pooled for the foregut and hindgut libraries respectively (SM107 and SM112, Figure  1B). By this stage endoderm patterning has initiated [14,35]. The notochordal plate at 0-6 somite stage and the notochord at 8-12 somite stage adjoin the DE and thus were included in the library [36].
A total of 322,208 tags were sequenced from these three longSAGE endoderm libraries [33]. Analysis of the three libraries revealed the expression of 54,093 different tagsequences (see Methods). There were 26,238 tagsequences present in the early endoderm library (SM108). Of these tag-sequences, 51% were unique to the early endoderm library as compared to the later endoderm libraries. Similarly, 25,097 and 25,509 tag-sequences were present in the foregut (SM107) and hindgut (SM112) libraries, respectively. In each of these libraries, approximately 50% of the tag-sequences were unique, compared to the other two endoderm libraries (Figure 2A). To determine which genes the tag-sequences represented, we first compared our tag-sequences to transcript databases (Refseq, MGC and Ensembl). Tag-sequences that did not correspond to annotated transcripts were then mapped to Ensembl gene units, which were extracted from the Ensembl database and include intronic regions and 1.0 kb upstream and downstream of annotated transcripts (Ensembl genes). Finally, tags were mapped to the mouse genome (UCSC). Of the combined 54,093 tag-sequences, 37% (19,782) mapped to known transcripts using the Refseq, MGC and Ensembl transcript databases, 12% (6,560) mapped to known genes using the Ensembl genes, implicating alternative splicing and alternative 3' UTRs of known genes, and 20% (10,954) mapped to the mouse genome. The remaining 31% (16,797) of the tagsequences did not map to any of these databases ( Figure  2B). Ninety percent of these unmapped tag-sequences were single tags, implying that many may have been generated by sequencing, PCR, or other errors. We have previously shown that many of these tag-sequences can be mapped by allowing a one-basepair mismatch, insertion or deletion [33]. However, some of these tag-sequences likely represent valid, novel transcripts, since 44 unmapped tag-sequences expressed in the endoderm were found at a level of at least 4 tags. For example, these 44 tag-sequences may span an unknown splice junction [37]. To simplify the analysis and validation in this study, we The total number of tags in each library is bracketed in the column heading.
focused on tag-sequences that unambiguously mapped to the most 3' position (position number +1) and the sense strand of the Refseq database (refer to Methods); 7,084 tag-sequences (13%) met these criteria.
Significantly, all of these endoderm genes were present in our endoderm libraries and excluded or present at low levels in the ectoderm libraries. The exception is Cxcr4, which although used as a DE marker, was expressed in both endoderm and ectoderm, reaffirming it as widely expressed [25]. Similarly, Sox2 is expressed in both ectoderm and endoderm libraries corresponding to published expression patterns [38]. All of the other ectoderm genes present in our ectoderm libraries were excluded or present at low level in the endoderm libraries. Overall, the expression patterns observed in our libraries supports known expression data for these genes, indicating that the libraries are representative of endoderm and ectoderm transcription.

Identification of foregut-specific genes
To identify genes that were specifically expressed in the foregut or the hindgut, a cross-comparison between the two libraries (SM107 and SM112, respectively) was performed. An initial list of genes was made by selecting tagsequences that were present at counts ≥4 for transcription factors (TFs) and signaling pathway components (SPCs), and counts ≥7 for other genes, in either the foregut or hindgut library. This threshold allowed us to identify the top 25 most highly expressed tag-sequences present exclusively in the foregut library and the top 20 most highly expressed tag-sequences present exclusively in the hindgut library, which was a tractable number for further validation [see Additional file 1]. By screening with both semiquantitative RT-PCR and quantitative RT-PCR, 14 of the 45 genes were shown to exhibit differential expression between the foregut and hindgut. Whole mount in situ hybridization was performed on these 14 genes. Six of these genes showed a ubiquitous expression pattern, making it difficult to determine whether there was differential expression within the DE. However, 8 genes did exhibit differential expression levels between the foregut and hindgut ( Figure 3). Seven of these genes, Trh, Otx2, Prrx2, Tbx1, Cyp26a1, Hoxb6, and Cdx1 were expressed in other tissues as well as endoderm at the early somite stage. Significantly, one of the genes, Pyy, was exclusively expressed in the foregut endoderm.

Expression of Pyy in the early mouse embryo
Pyy is known to be highly expressed in pancreatic islets and endocrine L cells of the lower gastrointestinal tract [39], but its early embryonic expression pattern has not been described. Due to the exclusive expression of Pyy in the DE at early somite stages from our analysis, we further examined Pyy expression pattern during early mouse embryogenesis. Whole mount in situ hybridization was performed on embryos collected from E6.0 to E9.5 stages ( Figure 4). Interestingly, Pyy was expressed in small lateral regions of the foregut DE as early as the 2 somite stage ( Figure 4A, 4B). At the 4 somite stage, the expression domains in the lateral region were expanded and a second expression domain in the medial ventral foregut was observed ( Figure 4C, 4D). Subsequently, the lateral expression domains expanded and extended anteriorly to Expression of Pyy in the early developing mouse embryo the medial ventral foregut, so that strong expression was observed in the lateral and ventral foregut at the 6-8 somite stages ( Figure 4E-J). Interestingly, the expression was restricted to the posterior half of the foregut and never observed in the anterior half of the foregut pocket. At early organogenesis stage, Pyy expression remained in the posterior foregut extending to the midgut junction ( Figure  4K-N). Thus, Pyy is expressed earlier than previously reported and demonstrates a dynamic expression pattern in the early DE.

Identification of novel genes expressed in the DE
In addition to identifying foregut-and hindgut-enriched DE markers, we wanted to identify additional novel genes with distinct expression patterns in the endoderm to facilitate DE patterning studies. Thus, to increase the efficiency of identification of novel endoderm genes, we chose to exploit the Mouse Atlas of Gene Expression Project database, which contained 133 libraries from different tissues and stages of development. We reasoned that if a gene was ubiquitously expressed, it would be present in most of the libraries. Conversely, if the expression of a gene were restricted to a specific cell-type, it would be present only in a specific subset of libraries. Indeed, by examining the expression patterns of our original list (foregut vs hindgut) of 45 tag-sequences in 133 longSAGE libraries generated by the Mouse Atlas Project, we discovered 9 genes that exhibited high tissue-specificity since they were present in only a few libraries ( Figure 5). Interestingly, 8 of the 9 genes demonstrating a tissue-restricted expression pattern matched the endoderm genes identified in our in situ hybridization analysis ( Figure 3). This suggests that in the context of looking for specificity of gene expression, the SAGE data is an excellent tool for identifying genes with tissue restricted expression.
To identify genes expressed in the DE, a second list was generated using tag-sequences present in the three endoderm libraries (7,084 tag sequences which were unambiguously mapped to the most 3' position and the sense strand of the Refseq database). We considered two factors, the total number of Mouse Atlas SAGE libraries in which a tag-sequence was present (L), and the total number of times that a tag-sequence was found in the three pooled endoderm libraries (T). We rationalized that higher T values and lower L values and thus higher T/L ratio would correspond to the degree of the endoderm-enrichment. We compiled a list consisting of tag-sequences with T>4 and L<58 and calculated the T/L ratio [see Additional file 2]. We removed the tag sequences whose T/L ratio was less than 0.21 to create a second list consisting of 60 genes. Confirming the effectiveness of these criteria, 6 out of the 60 genes were present in and validated by our first list, and 24 out of the 60 genes were previously shown to be expressed in endoderm, either with or without expression in other germ layer tissues, including Sox17, Foxa1-3, Ihh and Shh [see Additional file 2].
Of the remaining 30 genes, we successfully examined the expression of 26 genes using whole mount in situ hybridization. 21 of the 26 genes showed tissue-restricted expression patterns ( Figure 6, 7 and Table 2), while the remaining 5 genes showed ubiquitous expression at E8.5. Including the 6 candidate genes validated from the first list, the efficiency of our new screen for novel genes with tissue-restricted expression patterns was 84% (27/32). Interestingly, we found the majority of genes identified were not only expressed in the definitive endoderm, but also in other tissues such as yolk sac, ectoderm and mesoderm. We classified the 27 genes exhibiting tissuerestricted expression into five categories, based on their expression patterns ( Table 2). The first group includes two genes, 5730521E12Rik and Pyy, which were expressed exclusively in the DE. We described the Pyy expression pattern in the early mouse embryo above and, the 5730521E12Rik expression pattern is described below. Group 2 included genes that were expressed in the definitive endoderm and yolk sac endoderm, which support the functional similarity between these two lineages [6]. Group 3 contained genes that were expressed in the DE, yolk sac and another germ layer with a tissue-restricted pattern, and Group 4 contained genes that were not expressed in yolk sac and heart, but expressed in all 3 germ Expression of genes from the first candidate list within 133 mouse atlas SAGE libraries Figure 5 Expression of genes from the first candidate list within 133 mouse atlas SAGE libraries. Numbers on the X-axis depict each tag-sequence in the first candidate list, and the Y-axis depicts the number of the libraries in which a specific tagsequence is present. Thus a low bar reflects high tissue specificity, and vice versa. The 8 genes from the 1 st list exhibiting differential expression between foregut and hindgut by RT-qPCR as well as whole mount in situ hybridization have significantly lower bars. Names of these genes are provided on top of their respective bars. SPCs: Signaling Pathway Components; TFs: Transcription Factors.
layers. The genes in Group 5 were expressed in yolk sac endoderm at high levels, without obvious expression within the DE. The tags for these genes may be included in our libraries due to yolk sac endoderm contamination, which is difficult to avoid when collecting the DE tissue. Alternatively, these genes may be expressed in the DE at low levels but their expression in DE could be under-estimated by in situ hybridization due to very high levels in yolk sac endoderm. Thus most of the genes selected by our criteria for in situ hybridization showed complex tissuerestricted expression patterns in the early embryo, including the DE. Overall, these results indicate our approach was successful in the identification of novel markers of endoderm expression.

Expression of 5730521E12Rik in early mouse embryo
In addition to Pyy, 5730521E12Rik exhibited exclusive expression in the DE at early somite stages in our analysis. We further examined the expression pattern of 5730521E12Rik during early mouse embryogenesis. Interestingly, 5730521E12Rik was first expressed in a few cells at E7.25 in the endoderm at the posterior region adjacent to the embryonic-extraembryonic junction ( Figure 8A). At the late head-fold stage, 5730521E12Rik expression has expanded in the lateral region of the endoderm on the posterior side ( Figure 8B). As development proceeds, the bilateral expression domains extended anteriorly and medially and began to focus in the midgut region at early somite stages ( Figure 8C-F). By E9.0 strong expression was observed in the midgut ( Figure 8G). At E9.5, 5730521E12Rik expression was still maintained in the midgut region with the expression level decreased ( Figure  8H). Thus 5730521E12Rik expression was specific to the midgut during the gastrulation and early organogenesis stages in the mouse embryo.

Discussion
Formation, specification and patterning of the definitive endoderm are poorly understood in the embryo compared to other germ layers. Due to a lack of exploratory tools to aid these studies, interest in the identification of novel endoderm genes is growing. The recent enthusiasm for stem cell differentiation methodologies and the clinical potential for these cells have heightened the need for better tools and a further understanding of normal embryonic development. Several groups have undertaken largescale screens to identify novel genes that may be informative for developmental processes. In particular, in situ hybridization has been used to identify novel genes with unique expression patterns at mid-gestation (E9.5) [29]. However, while in situ hybridization is considered to be the ultimate and proven method to validate tissue-specific genes, obtaining embryos for in situ hybridization at appropriate stages is more costly and time-consuming in mouse than in chicken, frog or fish. Thus ensuring high efficiency in the screening for tissue-specific genes during mouse development is an important consideration.
To identify novel definitive endoderm specific genes, we used a longSAGE approach. We were able to enhance our screening efficiency since the endoderm longSAGE libraries were generated from enriched DE tissues at early stages of DE formation (E8.0-E8.5) that were obtained by a combination of proteolytic and manual micro-dissection methods. In addition, pre-selection of candidate genes by comparisons with 133 SAGE libraries from various tissues allowed us to eliminate ~95% of the widely expressing genes from our endoderm libraries. Overall, the efficiency of our screen for genes with DE expression was 69% (22/ 32, not including Group 5 genes which are highly expressed in the yolk sac). By including genes expressed in the yolk sac visceral endoderm, we observed 84% (27/32) of genes identified with endoderm expression. Significantly, two of these genes, Pyy and the Riken gene, 5730521E12Rik, were exclusively expressed in the DE lineage at early organogenesis stages of development.
Previous studies focusing on screening for novel endoderm genes have used cDNA cloning or microarray analysis [25][26][27][28][29]40]. Sousa-Nunes et al. identified 29/160 (18%) genes with restricted expression patterns from E7.5 mouse endoderm cDNA libraries using non-redundant sequence-based selection and in situ hybridization, but not all of these genes were endoderm enriched in their expression [40,41]. Sherwood et al. recently used fluorescent activated cell sorting to isolate definitive and visceral endoderm cell populations for microarray analysis [25]. They identified 18 out of 27 (67%) novel genes whose expression was enriched in endoderm. They defined a pan-endodermal signature composed of 22 novel and known genes that is preferentially expressed in definitive and visceral endoderm. Interestingly, neither study was able to identify novel genes that are expressed specifically in the DE.
The lack of DE specific genes may be due to sensitivity and depth of screening. Furthermore, the high functional similarity between visceral and definitive endoderm suggests that these tissues have highly related transcriptomes [25]. Several genes in our study were found to be expressed in both visceral and definitive endoderm, supporting the similarity of the two tissues. It is likely that some endoderm-specific or enriched genes were excluded from the gene list determined by our selection criteria. Our SAGE sampling depth (~100,000 tags per library) yields genedetection sensitivity approximately equivalent to that of fluorescence-based microarray approaches [42], and is thus sufficient for detection of abundant and moderately abundant transcripts but is likely insufficient for reliable detection of rare transcripts. Several previously known Tumor rejection antigen P1A 17 1 5 6 12 definitive endoderm, yolk sac and midbrain and tailbud

Group5: Genes expressed in yolk sac endoderm only
Tdh L-threonine dehydrogenase 19   foregut or hindgut markers were not present in our list likely due to their low expression level and/or their expression being restricted to few cells within the endoderm. For example, Prox1 is only expressed by the liver and pancreas progenitors beginning at the 7-8 somite stage [43]. Therefore, the Prox1 transcripts could be diluted by the total number of transcripts present in the foregut endoderm tissues, and thus not detected at our sequencing depth. With the advent of less expensive "next generation" sequencing this short-coming can be overcome by sequencing SAGE libraries to a greater depth. Furthermore, the foregut marker, Hhex, was missed in our analysis since it was expressed in 67 of our SAGE libraries, and thus did not fit our criteria for our validation lists (Table 1 and Additional file 2) [44]. Since many developmentally important genes are transcribed repeatedly and presumably function during multiple developmental processes, further refining of library and tissue choices for Histological sections through the embryos as indicated by the line in Figure 6, with arrows pointing to the DE staining Figure 7 Histological sections through the embryos as indicated by the line in Figure 6, with arrows pointing to the DE staining. Arrowheads indicate staining in visceral endoderm.
comparisons would be required to identify genes that are expressed in many stages of development and many tissues.
Pyy and 5730521E12Rik were identified to be exclusively expressed in the DE lineage at gastrulation and early organogenesis stages of development. Initially, Pyy is expressed at the early somite stage in the bilateral and medial regions of the foregut. Its early regionalized expression within DE reflects an early specification of cell fate along both the anterior-posterior and lateral-medial axis of the embryonic gut [6,45]. Subsequently, Pyy is expressed in the posterior foregut extending to the midgut junction, and at later stages (E14.5-adult) expression becomes restricted to the pancreas, stomach and intestine (data not shown). Interestingly, Tremblay et al. recently tracked progenitor domains in the anterior endoderm of mouse embryos, using vital dyes to label those cells at 1-10 somite stage. They identified two distinct types of DE progenitor cells, lateral and medial, arising from three spatially separated embryonic domains. These domains converge to generate the epithelial cells of the liver bud [13]. Intriguingly, the expression of Pyy follows a similar pattern as that observed by the lineage tracing of the liver bud progenitors. However, pancreatic progenitors were rarely labeled by this lineage tracing [13] suggesting that Pyy may not mark the identical domain. Deletion of Pyy in mice does not reveal any obvious defects in endoderm patterning [46,47]. However, genetic lineage tracing using Pyy-Cre and a ROSA26 reporter mouse strain demonstrated that in the adult, descendants of Pyy-expressing cells can contribute to the periphery of pancreatic islets and the L-type cells of the distal intestine [46]. The relationship between these later descendants and the early expression patterns has not been explored. Regardless, the dynamic expression pattern of Pyy appears to reiterate the morphogenetic movement of foregut progenitors along anterior-posterior and medial-lateral axes prior to tissue specification.
The RIKEN gene, 5730521E12Rik, expressed in the midgut region, is the first known gene that marks exclusively the entire midgut region at early organogenesis stages. Furthermore, 5730521E12Rik is the earliest DE specific and regional marker reported to date. Its early regionalized expression in the few cells in the posterior DE at as early as E7.25 embryo may reflect the early specification of the DE. Tam et al. recently depicted the sequential allocation and global pattern of movement of the DE in the mouse embryo during gastrulation, by tracing cells electroporated with Gfp or painted with carbocyanine dyes [45]. The observations from their study, together with previous fate mapping studies, suggested a probable sequence of allocation of the DE proceeding with (a) the most-posterior endoderm and the dorsal endoderm of the rostral segment of the foregut at early-streak stage; (b) the ventral endoderm of the rostral foregut and additional posterior endoderm at the mid-streak stage; (c) the dorsal and then the ventral endoderm of the posterior segment of the foregut at the late-streak to late-bud stage; and finally, (d) the endoderm of the embryonic mid-and hind-gut at the late-bud to early head-fold stage [45,[48][49][50][51]. Fascinatingly, the dynamic expression pattern of 5730521E12Rik suggests that it may mark the last population of the DE precursors recruited, thus is possibly a midgut lineage marker. Interestingly, 5730521E12Rik is identical to nephrocan (Nepn), which was recently identified by Mochida et al. as an inhibitor of Transforming Growth Factor-β signaling [52]. However, whether 5730521E12Rik plays an inhibitory role in vivo during endoderm formation and patterning needs to be further investigated.
The new endoderm genes we identified in this study will provide a valuable tool for further investigation into the underlying molecular mechanisms that regulate endoderm development. In particular, the dynamic expression patterns of Pyy, 5730521E12Rik and Trh from E6.5 to E9.5 provide intriguing insights into the endoderm fate mapping studies (Figure 4, 8 and unpublished data). In addition, Cpn1 and 1700011H14Rik showed strikingly similar expression patterns suggesting they may be co-regulated. Further expression and functional analysis of many of these genes will give insights into endoderm development. Moreover, these endoderm genes could be valuable markers to assess and optimize ES cell in vitro differentiation into endoderm and endoderm derivatives.

Conclusion
We identified novel as well as known genes that are expressed in DE progenitors by analyzing and validating DE longSAGE libraries. These genes provide a valuable tool for further investigation into the molecular mechanisms regulating endoderm development. Our study presents a successful application of analyzing and validating the large amount of data obtained by the Mouse Atlas of Gene Expression Project to identify tissue associated novel genes. The relatively high purity of the tissue source used for the construction of our DE longSAGE libraries and the comparison with a large number of longSAGE libraries from a variety of tissues and embryonic stages are the two critical factors for achieving an efficient screen.

Tissue collection and generation of SAGE libraries
Obtaining enriched DE tissue was achieved by a combination of proteolytic and manual micro-dissection methods [14]. E8.0-E8.5 embryos were isolated from timed pregnant female C57BL/6J mice. After removing the extraembryonic membranes, the embryos were transferred to 1% trypsin/Hanks and incubated for 30 minutes on ice. Then 0.01%DNase/20% FBS/Hanks were added to block the activity of trypsin and to digest genomic DNA to reduce stickiness of the tissue. Next, using polished tungsten needles or fine tip forceps, endoderm and ectoderm were separated to minimize mesoderm contamination, and then transferred into Trizol (Invitrogen) ( Figure 1A). Somite 0-6 endoderm pieces were pooled for the early whole endoderm library. Somite 8-12 endodermal portions were divided into foregut and hindgut region, and then pooled for foregut and hindgut libraries respectively ( Figure 1B). In total, 3.5 µg, 2.7 µg and 3.5 µg total RNA were isolated from 124 early whole endoderm, 110 foregut and 115 hindgut pieces respectively. RNA quality was assessed using a Bioanalyzer (Agilent). Each SAGE library was constructed with 2.5 µg of DNA-free total RNA using the Invitrogen I-SAGE Long kit and protocol as previously described [32,33].

SAGE data analysis
SAGE data was analyzed using DiscoverySpace software [53]. All SAGE libraries were generated by the Mouse Atlas of Gene Expression project [33]. They were filtered for sequence quality using a 95% quality cut off for all tags. Tag to gene mapping was performed using the mouse Refseq, MGC and Ensembl databases using the CMOST plugin in DiscoverySpace. Tags were considered sense position matches if they mapped in the same 5' to 3' orientation as the gene, and antisense matches if they mapped in the opposite orientation. Tag 'position' was determined by sequentially numbering NlaIII restriction sites from the 3'-most end (position 1) onward (i.e. next 5' tag would be position 2, and so on). A tag was considered unambiguous if it mapped to a single gene in a sense position and ambiguous if it mapped to multiple genes in a sense position.

RT-PCR, whole mount in situ hybridization and histology
Semi-quantitative RT-PCR followed standard protocols. An ABI 7900 real-time PCR system (Applied Biosystems) and SYBR Green supermix (Applied Biosystems) were used for quantitative real time PCR. RNA from each tissue was prepared using Trizol (Invitrogen). Triplicate cDNAs were obtained by reverse transcription of 1 µg of total RNA from newly isolated batches of endoderm tissue. The primers used in the semi-quantitative RT-PCR and realtime PCR are listed in Additional file 3.
Whole mount in situ hybridization was performed as described previously [54]. Probe templates were generated by RT-PCR amplification from total RNAs isolated from E8.0-E8.5 endoderm, with average sizes of 400-800 bp, followed by sequence verification [see Additional file 4]. At least three embryos at each stage were examined for each probe, and restricted expression patterns were con-firmed by independent sets of hybridizations. After whole mount in situ hybridization and photographing, the embryos were embedded by standard procedures in paraffin, sectioned at 8 µm, dried overnight, dewaxed in xylenes, and mounted for imaging.