The pancreas is composed of two types of tissue: exocrine and endocrine. The exocrine pancreas is made of acinar cells and secretes digestive enzymes into a network of ducts, while the endocrine pancreas consists of the islets of Langerhans and secretes hormones into the bloodstream. Pancreatic β cells are highly specialized cells within the islets of Langerhans responsible for producing vast amounts of insulin in response to changing glucose levels in blood. β cells are affected during Type-1 Diabetes (T1D) and Type-2 Diabetes (T2D) and are a major focal point of researchers in both fields. Availability of a complete list of transcripts expressed in human β cells, along with the transcriptomes of other cell types in endocrine and exocrine pancreas will aid T1D and T2D research.

Microarray technology is presently the preferred method for global (comprehensive) gene expression measurement and has been applied successfully to pancreatic islet and β cell-focused studies from human, mouse and rat 1234. MPSS is an alternative technology that estimates gene expression by counting short sequence signatures generated from up to one million expressed sequences per run. MPSS analyses provide very deep transcriptome analyses of individual tissues or cell types 5. Unlike microarrays, MPSS eliminates the need to predefine genes that can be detected and samples the transcriptome deeply enough to detect transcripts expressed at levels as low as three copies per cell 6.

Systems biology is a multi-disciplinary science that seeks to quantify the molecular elements of a biological system, determine their interactions, integrate these data into molecular network models and then correlate network dynamics (changes in the components and architecture of the network) with developmental, physiological and pathological behaviors 7. Such dynamic models serve to generate predictive hypotheses that can be experimentally verified. A first step toward constructing a systems biology network model is to build a comprehensive quantitative expressed-mRNA database reflecting dynamically changing transcriptomes of the cell types of interest (at different stages of their development, functional operation or disease progression). There are two types of dynamic molecular networks that in practice are closely integrated: protein and gene regulatory networks. Protein networks (protein/protein/small molecule interactions), for example, transmit information from the cell surface to the nucleus, mediate metabolism and provide the cell with structural integrity. On the other hand, gene regulatory networks integrate/modulate information and control behavior of protein networks or complex molecular machines through the action of transcription factors. Hence, delineation of the expression patterns of transcription factors of a particular cell type provides the components of its gene regulatory networks and initial insights into the networks that mediate its functional regulation. Specific changes observed in these networks under diseased states might serve as biomarkers of disease progression. Moreover, specific expression patterns, static or temporal, of a gene can provide important clues to its physiological function.

Current efforts to catalog gene expression in tissues lack detailed information about pancreas and pancreatic islets. Symatlas, the most widely used database of tissue expression, contains human and mouse expression data from up to 79 tissues under basal conditions 8. In Symatlas, pancreas is represented as only one column without disclosing differences at the cellular level. The same is true for other complex tissues, such as liver and lung. Nor is an effort made to compare differences and commonalities between the same tissue in human and mouse. Having more detailed atlases that attempt to represent information from multicellular organs across different species and different data types is warranted.

There are several technical challenges that preclude obtaining of pure human β cells. These include limited availability of human material, no established protocol to isolate β cells from human pancreatic islets and the lack of human beta cell lines. In order to obtain a relatively complete β cell gene expression profile, we performed MPSS analysis of two independent human islet samples. Human islets contain α and β cells in a ratio between 0.3–0.5 to 1 and make up approximately 80% of the whole cell population, that are difficult to separate in human samples. Accordingly, we also characterized by microarray analysis rat β and α cells that are readily separated by FACS. The assumption is that the islet-cell transcriptomes will be similar in human and rat (and mouse). We also performed microarray analyses on the rat INS-1E cells, a glucose-responsive insulinoma cell line commonly used as a model to study β cell biology. We combined these four new types of data with expression data from the literature and public databases on pancreas and insulin-producing cells. The resulting 'β Cell Gene Atlas' (BCGA) incorporates basal gene expression data from experiments performed with pancreatic β and other cells in human, rat and mouse. The data assembled in the BCGA provide a major resource contribution to our understanding of β cells and the expression landscapes of cells in the pancreas that will benefit efforts to understand and ultimately prevent diabetes.

We have used both MPSS and microarray data generated in the current study, together with publicly available expression data, to build a repository of β cell and pancreas gene expression, the β cell Gene Atlas (BCGA). The BCGA contains expression information for genes expressed particularly in β cells and other cell types in pancreas of humans, rats and mice. Obtaining a complete list of genes expressed in a certain cell type is the first step in generating a model of the biological networks that are active under normal conditions but perturbed under disease. Therefore, the Atlas is an essential tool that will allow application of systems biology to the field of pancreatic islet research and T1D research in particular.

BCGA can be queried with virtually any public identifier (Gene symbols, Entrez Gene ids, ENSEMBL ids) from all three species. Results are returned for all species that is orthologous to the queried gene. BCGA is a resource that allows side by side comparison of microarray data related to pancreas with a special emphasis on pancreatic β cells. BCGA will be useful in several research areas: β cell-specific biomarker discovery (as potential candidates for early diagnosis, determining the stages of disease progression and response to therapy), β cell regeneration (as a benchmark to compare regenerated islets) and pancreas development projects (as a list of genes (or transcripts) expressed in functional β (and α) cells encoding protein molecular machines, protein interaction networks and transcription factors encoding gene regulatory networks).

Microarrays are presently the standard method for global gene expression studies. However, the list of genes represented on microarrays is based on a priori knowledge and represents a subset of the whole genome. Compared to microarrays, MPSS data has the advantage of being quantitative and permits comparison of transcript copies between different datasets. MPSS analysis of two human islet samples resulted in detection of around 7,600 transcripts. These genes include many highly characterized genes as well as genes whose functions in islets are unknown. Results from microarray and MPSS platforms agree to a certain extent, while there are some genes that are only detected by one method. Similar moderate agreement between the two platforms has been observed in other studies 3334. MPSS technology is unable to detect approximately 7% of known genes lacking a DpnII restriction enzyme site 30. The present MPSS data do not include 630 genes that are detected in the upper 90^th percentile levels in the human islet microarray data, which contains around 130 genes that lack a DpnII site. Therefore, this is not enough to explain the observed discrepancies between the two technologies. Microarray probes may have undesired properties, such as cross-hybridization, that could lead to false positives that are not detected by MPSS. The two methods have differential bias, including those related to G+C content of gene sequences 35. The effect of G+C content on the stability of hybridized sequences is well known, with higher G+C content corresponding to more stable DNA duplexes in microarray platform. On the other hand, the G+C content might interfere with the sequencing properties in MPSS data. Another possible source of error is the annotation of the MPSS signatures. Around 20% of the signatures in our islet dataset are mapped to more than one gene (i.e. the signatures align to more than one location along the human genome) and these were discarded from the analysis (Supplementary Information). There are also many signatures not converted to genes. The results of Encyclopedia of DNA Elements (ENCODE) project that focuses on functional elements of 1% of the human genome, has recently been published 36. The majority (>60%) of interrogated loci presented potential new exons mapping in their introns, while two-thirds (68%) of the investigated loci showed potential new putative TSS upstream of their annotated first exon, often reaching into neighboring genes 36. Current human genome annotations probably fail to determine these yet to be identified exons and transcripts. Furthermore, some signatures (2209) could not be aligned to the genome during our annotation process. In subsequent analysis, approximately 1000 signatures could be mapped to mitochondrial genome. We have been able to align a subset of the unannotated signatures to transcripts in NCBI Trace Archive database http://www.ncbi.nlm.nih.gov/Traces/trace.cgi. These signatures are probably from genes residing in regions that could not be assembled to the rest of human genome. This might indicate widespread existence of unknown exons and/or genes that remain to be identified. Finally, polymorphisms probably contribute to the high number of transcripts detected as present in the BCGA. The human islet microarray data in the atlas originate from 20 different individual samples and 9 experiments while the MPSS data are derived from only two individuals.

The variability in the MPSS results (see Results) might be explained by the differences in the depth of sequencing (1.05 million vs 1.31 million signatures) and is in line with previous observations of the variability between individual donors 37. It also could arise from cells being at different stages of physiological response. One also must note that there may be heterogeneity in the cell populations in the pancreas that the ratios of these different cell types may differ in different human islet preparations. Limited availability and prohibitively expensive cost of MPSS precluded us from performing more MPSS experiments. However, the cost of signature sequencing based experiments (such as Solexa, ABI) has dropped dramatically. We plan to carry out more experiments of this type of next-gen sequencing and add to BCGA.

Transcription factors are essential for maintaining the gene regulatory networks that regulate all cellular functions: knowledge of regulatory behavior is key to predictive models of biological systems 38. In the present study we have identified 346 and 312 transcription factors that are expressed in rat β and α cells respectively. Transcription factors with HLH, HD, bZip, NR and LIM domains make up around 70% of the basally and differentially expressed factors in rat β and α cells [see Additional File 4]. Complexity of bZip transcription factors functioning is achieved through dimerization and high throughput analysis of bZip-bZip interactions detected 200 preferential binding sites that regulate distinct genes 39. Atf-3 and binding partners Fos and Jun are enriched in α cells while interacting partners Atf-6 and Xbp-1 are enriched in β cells [see Additional File 4]. Atf-6 and Xbp-1 are involved in initial stages of the unfolded protein response and higher expression of these genes might be indicative of a stronger unfolded protein response in β cells compared to α cells, as suggested by previous studies 40.

The quantitative nature of MPSS data and its digital counting of individual mRNA transcripts allow direct comparison of these datasets without further normalization. Comparison of MPSS expression in human islets to other tissues indicates that 940 genes are expressed at >80% specificity in the pancreatic islet cells. 324 out of 940 islet specific genes are enriched in β cells compared to α cells. Among these genes are well-known β cell-specific genes such as PDX1, NEUROD1, IAPP, TCF1, NKX6-1 and ICA-512 [see Additional File 7]. Massive β cell death takes place during T1D and to a certain extent in T2D, raising the possibility to search for β cell specific blood markers secreted or released by β cell-specific genes. It is expected that the levels of these β cell-specific blood proteins will reflect the operation of the networks present in these cells, which should allow us to distinguish between normal and diseased β cells. Moreover, new blood proteins may be released from β cells undergoing apoptosis thus providing an indication of the extent of β cell destruction. Accordingly, blood from healthy and recently diagnosed people could be monitored to follow β cell mass and network perturbation during disease progression. The results of the present study, and the newly constructed BCGA, provide a list of potential candidates for this novel approach.

It is worth mentioning that the information contained in the BCGA is affected by the limitations inherent to microarray platforms and the algorithms utilized for homologue detection. For example, there are currently no tools in the BCGA to match genes belonging to the same family across species. We expect to improve these aspects in the future releases.

The BCGA is a repository that integrates high-throughput gene expression datasets obtained in untreated β cells and other cells of the pancreas. The available datasets include the rat β and α cells microarray data performed in this study, MPSS data from human pancreatic islets as well as other publicly available dataset. BCGA contains gene abundance estimates for each gene in different cell types across three species. It is well known that gene networks are dynamic and change during cell cycle and physiological/pathophysiological responses. Our BCGA data provides an initial picture of the status of these networks and levels of gene expression under basal conditions. These data enable interesting comparisons between different cells and cell lines in human, rat and mouse pancreas. Our analysis using different models of β cell expansion or reduction suggests that BCGA contains cell-specific data. This information will be valuable for future studies involving expression changes and dynamic processes. The presently developed data pipeline is set up to incorporate any new microarray data or other platforms, such as proteomics, and to make it available to the research community promptly on T1DBase. The atlas is available at T1DBase http://T1DBase.org/page/AtlasHome, a bioinformatics resource for T1D researchers 41. This cumulative information, coupled to pathway analysis, will be of great value for the ultimate understanding of gene/protein networks regulating β cell development, function and death.

BCGA: β cell Gene Atlas; MPSS: Massively Parallel Signature Sequencing; Tpm: transcripts per million; FDR: False Discovery Rate.

The authors declare that they have no competing interests.

BK, D. Baxter, JR, DF and NW performed experiments, BK analyzed data, BK, D. Burdick and NG wrote software, BK, NG, DE and LH conceived of the study, and participated in its design and coordination. DE and NG helped to draft the manuscript. All authors read and approved the final manuscript.