Browse Subject Areas

Click through the PLOS taxonomy to find articles in your field.

For more information about PLOS Subject Areas, click here.

  • Loading metrics

Gene expression signature predicts human islet integrity and transplant functionality in diabetic mice

  • Sunil M. Kurian ,

    Contributed equally to this work with: Sunil M. Kurian, Kevin Ferreri

    Roles Conceptualization, Data curation, Formal analysis, Methodology, Writing – original draft, Writing – review & editing

    Affiliation Department of Molecular and Experimental Medicine, The Scripps Research Institute, La Jolla, California, United States of America

  • Kevin Ferreri ,

    Contributed equally to this work with: Sunil M. Kurian, Kevin Ferreri

    Roles Conceptualization, Data curation, Formal analysis, Investigation, Methodology, Project administration, Software, Supervision, Validation, Writing – original draft, Writing – review & editing

    Affiliation Department of Translational Research and Cellular Therapeutics, Diabetes, and Metabolism Research Institute, City of Hope National Medical Center, Duarte, California, United States of America

  • Chia-Hao Wang,

    Roles Data curation, Formal analysis

    Affiliation Department of Translational Research and Cellular Therapeutics, Diabetes, and Metabolism Research Institute, City of Hope National Medical Center, Duarte, California, United States of America

  • Ivan Todorov,

    Roles Formal analysis, Investigation, Methodology

    Affiliation Department of Translational Research and Cellular Therapeutics, Diabetes, and Metabolism Research Institute, City of Hope National Medical Center, Duarte, California, United States of America

  • Ismail H. Al-Abdullah,

    Roles Methodology

    Affiliation Department of Translational Research and Cellular Therapeutics, Diabetes, and Metabolism Research Institute, City of Hope National Medical Center, Duarte, California, United States of America

  • Jeffrey Rawson,

    Roles Data curation, Methodology

    Affiliation Department of Translational Research and Cellular Therapeutics, Diabetes, and Metabolism Research Institute, City of Hope National Medical Center, Duarte, California, United States of America

  • Yoko Mullen,

    Roles Supervision

    Affiliation Department of Translational Research and Cellular Therapeutics, Diabetes, and Metabolism Research Institute, City of Hope National Medical Center, Duarte, California, United States of America

  • Daniel R. Salomon,

    Roles Conceptualization, Resources, Supervision, Writing – original draft, Writing – review & editing

    Affiliation Department of Molecular and Experimental Medicine, The Scripps Research Institute, La Jolla, California, United States of America

  • Fouad Kandeel

    Roles Conceptualization, Formal analysis, Supervision, Writing – original draft, Writing – review & editing

    Affiliation Department of Translational Research and Cellular Therapeutics, Diabetes, and Metabolism Research Institute, City of Hope National Medical Center, Duarte, California, United States of America

Gene expression signature predicts human islet integrity and transplant functionality in diabetic mice

  • Sunil M. Kurian, 
  • Kevin Ferreri, 
  • Chia-Hao Wang, 
  • Ivan Todorov, 
  • Ismail H. Al-Abdullah, 
  • Jeffrey Rawson, 
  • Yoko Mullen, 
  • Daniel R. Salomon, 
  • Fouad Kandeel


There is growing evidence that transplantation of cadaveric human islets is an effective therapy for type 1 diabetes. However, gauging the suitability of islet samples for clinical use remains a challenge. We hypothesized that islet quality is reflected in the expression of specific genes. Therefore, gene expression in 59 human islet preparations was analyzed and correlated with diabetes reversal after transplantation in diabetic mice. Analysis yielded 262 differentially expressed probesets, which together predict islet quality with 83% accuracy. Pathway analysis revealed that failing islet preparations activated inflammatory pathways, while functional islets showed increased regeneration pathway gene expression. Gene expression associated with apoptosis and oxygen consumption showed little overlap with each other or with the 262 probeset classifier, indicating that the three tests are measuring different aspects of islet cell biology. A subset of 36 probesets surpassed the predictive accuracy of the entire set for reversal of diabetes, and was further reduced by logistic regression to sets of 14 and 5 without losing accuracy. These genes were further validated with an independent cohort of 16 samples. We believe this limited number of gene classifiers in combination with other tests may provide complementary verification of islet quality prior to their clinical use.


The pathophysiology of Type 1 Diabetes Mellitus (T1DM) is the result of autoimmune destruction of insulin-producing beta cells in the pancreas. Several immunotherapy strategies are suggested in order to reduce the immune mediated destruction of the insulin producing cell [1,2]. In addition, a promising treatment paradigm for T1DM is replacement of the missing beta cells with islet cells isolated from allogeneic donor organs [3,4]. Successful islet transplantation has been shown to improve glycemic control, induce insulin independence or significantly reduce insulin requirements and, most importantly, provide several years of freedom from life-threatening hypoglycemic episodes [57]. Although concerted efforts from several groups have resulted in progress in the field over the last decade, transplantation outcomes have not been consistent between the various transplant centers [8]. In addition to problems with alloimmune rejection and residual auto-immunity directed against the islet graft, the ability of human islet isolation centers to consistently provide viable and functional islet cells varies widely within and especially between transplant centers [8,9]. This is confounded by the lack of robust, reproducible and standardized methods for gauging the suitability of specific islet preparations for clinical transplantation [1012].

Consequently, a major effort in the field has been the development of methods for evaluating islets prior to clinical transplantation which are predictive of outcomes in the patients. Currently, the best evidence of islet function is reversal of diabetes by transplantation of human islets into diabetic immunocompromised mice [13]. However, the assay requires several weeks to obtain results and is therefore not suitable for assessment of the cells prior to transplantation, which typically occurs within three days post-isolation. As a result, research efforts have focused on the identification of surrogate parameters that are predictive of islet graft function and which can be evaluated within the relatively short time between islet isolation and infusion into the patient.

Our group has investigated the use of percent beta cell apoptosis (BAP) and glucose-responsive oxygen consumption rates (OCR) as predictors of islet graft function. Each of these approaches independently predicts reversal of diabetes in mice with reasonable accuracy (0.856 for BAP [14] and 0.793 for OCR [15]). Furthermore, these methods are rapid enough to obtain results prior to clinical use of the islet preparations. We also demonstrated that OCR provides identical results independent of the institute performing the assay. However, these widely used in vitro approaches focus solely on the immediate integrity of the islet preparation without regard to potential for in vivo islet function or graft-host interaction, elements more likely to be important for long-term efficacy following transplantation.

In considering the factors that make an islet preparation “good” for clinical use, we speculated that both the function of the islet preparation and the interaction with the recipient would be governed by the expression of specific islet genes. Therefore good islet preparations would have a distinctive “gene signature”. To test this hypothesis, whole genome RNA expression analysis using microarrays was performed on 59 human islet preparations in parallel with assessment of islet function by transplantation into diabetic mice. Using this approach, a set of 262 microarray probesets representing 199 human genes was associated with the ability of islets to reverse diabetes in mice. These probesets were able to predict the outcome of transplantation studies with an accuracy of over 83%, suggesting that a “gene signature” could be associated with islet quality.

Importantly, the gene classifiers were functionally associated with islet biology and were predominantly associated with inflammation and repair mechanisms rather than metabolic function. Interestingly, the gene signature showed little overlap with gene expression profiles associated with our other measures of islet quality, BAP and OCR, suggesting these islet quality tests measure different aspects of islet biology. Finally, we demonstrate that the microarray-based gene signature assessment is readily adaptable to rapid evaluation of islet preparations using a PCR based methodology. In summary, our data demonstrate the feasibility of using islet gene expression as a metric for functional islet quality assessment in the context of clinical cell therapy programs.


Islet gene signature correlated with reversal of diabetes

To identify a gene signature associated with islet quality, each islet preparation was assigned to one of two classes based on their ability to reverse diabetes, namely good islets, which resulted in reversal of diabetes after transplantation into diabetic mice, and bad islets, those which failed to reverse diabetes (see Materials and methods for criteria). To minimize bias in the analysis, the 59 samples were randomly assigned three times into two groups, a training set (Group 1) and a validation set (Group 2), while maintaining approximately equal numbers of good and bad samples in each group. In each iteration Group 1 was used to identify microarray probesets representing individual genes that were associated with either good or bad islet preparations, then Group 2 was used to test each of the resultant probesets for the ability to correctly predict the category (good or bad) of each islet preparation (see Materials and methods for a detailed description). Probesets that had 100% cross-validation efficiency (%CV; i.e. effectiveness at classifying the samples correctly) were collected as a Predictor Classifiers list. The combined Predictor Classifiers from the three randomizations yielded a total of 262 unique probeset classifiers for islet quality representing 199 genes that had 100% predictive accuracy (Table 1). The data showed that 135 of the 262 probesets (51.5%) exhibited higher expression levels in bad islet preparations, and 127 probesets had higher expression in good islet preparations.

Table 1. Probeset classifiers for reversal of diabetes arranged by p-value.

The 36 classifier subset is highlighted in BOLD. P-values are the Parametric P-values obtained during the analyses of all the microarray datasets. Fold-change is the ratio of average expression (intensity level) in the Bad samples divided by average expression in the Good samples. Probes are the identification numbers of the Affymetrix U133 Plus 2.0 GeneChip probesets.

The predictive value of the combined 262 probeset was then tested by supervised clustering of the 59 islet preparations. Cluster analysis generated two distinct clusters (Fig 1A) with the majority of bad islet preparations in Cluster 1 and majority of good preparations in Cluster 2. Only three bad islet preparations were misclassified resulting in an 89% predictive accuracy for the good preparations. However, ten of the total 36 good samples were misclassified as bad (27%). Notably, seven of these ten misclassified good preparations were in a small sub-cluster adjacent to another sub-cluster containing 2 of the 3 misclassified bad preparations, suggesting the possibility that these smaller clusters on both sides of the line of separation represent a class of intermediate quality islets. Overall however, these data demonstrate that using the consensus set of 262 probesets as a predictor would result in the transplantation of very few bad islet preparations.

Fig 1. The 262 and 36 probeset lists predict islet quality.

(A) 59 human islet preparations were clustered using the 262 probeset classifier. Eighty-nine percent (25 of 28 samples) of the good islets (green) clustered within the same quality class, and 68% (21 of 31 samples) of bad preparations (red) clustered together. The overall predictive accuracy of this classifier set was 83%. The heat map depicts expression level of the 36 probesets in each sample; Red: probeset with higher expression in bad islets; Green: probeset with higher expression in good islets; the intensity corresponds to the fold-difference in gene expression. (B) The 262 classifier set (diamonds) and the 36 classifier (squares) were analyzed by ROC curve analysis for their ability to discriminate between the good and bad classes of islet preparations. Both classifiers perform better at identifying poor islet preparations (NPV) than effective preparations (PPV).

Nevertheless, expression analysis of such a large number of genes may be difficult to implement in a clinical transplantation program on a routine basis, therefore we investigated ways to reduce the number of classifiers without losing predictive power. The Predictor Classifier lists from the three randomizations were compared and it was observed that they shared 36 “core” classifiers. The 36 classifiers clustered the samples into two groups with a distinct gene expression pattern for each class of samples (Fig 1A). The clusters predicted by the 36 probeset list were identical to the groups predicted by the 262 probeset list and also had an overall predictive value of 83%. Receiver operating characteristic analysis showed that the curves of two classifier sets (Fig 1B) were nearly identical, indicating that the reduced set of classifiers had equivalent predictive accuracy to the larger set, and that it correctly identified 79% of the good islet preparations (PPV; positive predictive value) and 86% of the bad preparations (NPV; negative predictive value).

Further analysis of the 36 probeset classifier was done using ANOVA gene expression model (Partek Genomics Suite) to classify each sample in each of six randomized sample groups using the probesets to determine its ability to predict islet function in vivo. The correct classification (good or bad) of each of the six random groups of samples ranged from 82% to 90.5% with an average of 85.3% correct prediction of the outcome in mice. In summary, both the 262 probeset classifier and the reduced list of 36 probesets, representing just 25 genes, predicted post-transplantation islet function with comparable and high accuracy.

Refinement of predictor list by logistic regression

To further assess redundancy in the 36 probeset above that is necessary for the successful classification of islets into good and bad preparations, logistic regression analyses were conducted. The results showed considerable redundancy even among the 36 probesets, with fourteen probesets in the initial full model (EST2, KCNMA1, EST5, PKIB, EHD4, SEPT9, MIR181A2, RND3, PMEPA1, IFITM2, CARD6, MNX1, RNF187, MAPT) showing excellent separation between good and bad islets, with a maximum 0.96 true positive rate and a zero false positive rate (Fig 2A and 2B). Interestingly, using further step-wise model simplification, the number of gene probes can be reduced down to five (EST2, KCNMA1, RND3, PMEPA1, CARD6) while maintaining a maximum true positive rate of 0.93 and a false positive rate of zero (Fig 2C and 2D).

Fig 2. Reduction to 14 and 5 probeset lists by logistic regression.

(A) Boxplot and (B) ROC curve for the fourteen probesets model. The fourteen probesets are EST2, KCNMA1, EST5, PKIB, EHD4, SEPT9, MIR181A2, RND3, PMEPA1, IFITM2, CARD6, MNX1, RNF187, MAPT. The model achieves a 0.96 true positive rate and a zero false positive rate at score threshold value of 0.5. (C) Boxplot and (D) ROC curve for the five probesets model. The five probesets are EST2, KCNMA1, RND3, PMEPA1, CARD6. The model achieves a 0.93 true positive rate and a zero false positive rate at score threshold value of 1.2.

Classifier gene function

Examination of the 262 probesets (Table 1) revealed that approximately half (135 of 262) were more highly expressed in bad islet preparations while the other half were higher in good islet preparations, suggesting that the difference observed in in vivo function was not solely due to up-regulation of deleterious molecules, but also to the preservation or up-regulation of beneficial molecules. Further investigation of gene function by Ingenuity Pathway Analysis revealed that two apparent processes were competing in the islets to affect their in vivo function (Fig 3A). The most significant functional network was Endocrine System Development, which along with certain less significant networks (Cancer, Organ Development, Cellular Growth and Proliferation, Cardiovascular System Development, and Tissue Development), indicates an increase in tissue repair mechanisms such as cell proliferation and cell differentiation. Of the genes associated with endocrine development, 73% exhibit higher expression in good islet preparations, and together these genes form a pathway associated with beta cell development (Fig 3B). Increased expression of cell growth and differentiation pathways implies that ongoing islet repair is associated with better in vivo function.

Fig 3. Functional pathways of the 262 classifier set.

(A) Ingenuity Pathway Analysis was used to group the classifiers for reversal of diabetes into functional pathways. The twelve most significant functional pathways are listed in order of–log(p-value) with the most significant pathway (Endocrine System Development; p = 2.60 x 10−9) at the top. The number of probesets associated with each pathway is also listed. (B) Pancreatic endocrine cell development and regeneration pathway showing genes identified in the 262 classifier list (colored). (C) The network of inflammatory and immune related molecules predominantly expressed in non-functional islets prior to transplantation. Green means higher expression in good preparations, Red means higher expression in bad preparations, and the intensity corresponds to the fold-difference in gene expression. The pathway figures are adapted from Ingenuity Pathway analysis and the KEGG pathway database.

The second most significant functional network was Inflammatory Disease (Fig 3A). Along with other networks (Hepatic System Disease, Gastrointestinal Disease, Neurologic Disease, and Inflammatory Response), these classifiers form an interconnected network of molecules associated with the innate immune system and the inflammatory response (Fig 3C). Importantly, 69% of the genes in Inflammatory Disease have higher expression in bad islets, suggesting that beta cells may participate in their own dysfunction by up-regulation of pro-inflammatory molecules even prior to transplantation. Some of these, such as CCL2 (also known as MCP1), have been previously reported to have a negative effect on islet function [1619]. However, most of the genes in these pathways have not been studied in relation to islet biology or diabetes.

Gene signatures associated with oxygen consumption rates and apoptosis

We previously reported that both the percentage of apoptotic beta cells (BAP) [14] and glucose-responsive oxygen consumption rates (OCR) [15] individually provided reasonable predictive accuracies of subsequent islet graft function of 0.86 (95% confidence interval: 0.75–0.96) and 0.79 (95% confidence interval: 0.61–0.97), respectively. OCR and BAP results represent different aspects of the islet preparation, with OCR reflecting the metabolic responsiveness of the islets and BAP the viability of the beta cell population. To determine the extent to which genes responsible for these characteristics were shared with each other and with the 262 probeset classifier, islet gene expression data was analyzed to identify genes associated with OCR and BAP results. To obtain an accurate gene expression profile for these characteristics using the current microarray datasets, transplantation results of current samples were used to set the thresholds of each method. For this sample set, the thresholds for OCR good and bad islets were OCR >0.191 and <0.085 nmol O2/min/100 islets, respectively. Similarly, good and bad islets had BAP <1.91% and >4.30%, respectively.

The gene expression data were analyzed by class comparison based on these thresholds to define good and bad classes of islet preparations. This yielded a set of 985 probesets, representing 736 genes, that discriminated between high and low OCR, and a set of 1056 probesets, representing 790 genes that differed between high and low BAP. Pathway analysis revealed the OCR and BAP gene sets represented two strikingly different classes of genes functionally. Good islets, as defined by OCR, showed 16 significant pathways (p<0.005), a majority (65%) of which were associated with metabolism (Fig 4A). By contrast, the 44 significant pathways representative of good islets classified by BAP expressed greater association with signaling pathways, including key pathways related to islet biology, such as mTOR and AMPK signaling, and only two metabolic pathways (5%). These results suggest that OCR and BAP measured two distinct aspects of islet biology, as represented by transcriptional profiling. Moreover, comparison of the probeset classifiers derived from all three functional assays (reversal of diabetes in mice, OCR, and BAP) showed that the majority of genes associated with each metric was unique (Fig 4B).

Fig 4. Canonical pathways of the OCR and BAP classifier sets.

The results of Ingenuity Pathway Analysis of the probeset classifier lists associated with OCR and BAP. The top ten pathways associated with OCR (green) and BAP (red) are listed in order of the level of statistical significance (-log(p-value)). Pathways associated with OCR are mostly metabolic, whereas those associated with BAP are various signaling pathways. B. Venn diagram of the probeset classifier lists from diabetic mouse data (Transplant), OCR results, and BAP shows that there is overlap of only 9 probesets. This indicates that the three parameters are measuring distinct characteristics of islet function, which is supported by the diversity of functional pathways associated with each parameter. The diagram was created using the VENNY website tool [20].

Development and verification of a diagnostic test for islet quality

Implementation of a “gene signature” for gauging islet quality within a clinical islet transplantation program requires a rapid, inexpensive, and reproducible method of measuring differential gene expression such as quantitative RT-PCR (qRT-PCR). To test this approach, an independent cohort of 16 new islet preparations comprising 8 good samples and 8 bad were analyzed by qRT-PCR for differential expression of a subset of the 36 probeset classifier for reversal of diabetes. The 36 classifiers represent 25 known genes, and of these there were twenty expression assays readily available on the qRT-PCR-based OpenArray TaqMan platform. These twenty genes represent 25 (69%) of the 36 probesets. The results show that expression of 10 of the 20 genes are significantly different (p<0.05) between islets that reverse diabetes and those that do not (Table 2). Analysis of these 10 significant genes as a single metric by two-way ANOVA proves highly significant (p<0.0001) for islet quality, and together have a predictive accuracy for reversal of diabetes of 86% (ROC analysis; Area Under the Curve = 0.8640 ± 0.0285). These results demonstrate that qRT-PCR can be a useful method for assessing expression levels of the gene classifiers and that the expression levels of this subset of 10 genes successfully predicts in vivo outcome in the diabetic mouse model.

Comparing the results of the limited 14 probeset with those of the qRT-PCR revealed that 5 (KCNMA1, SEPT9, RND3, IFITM2, CARD6) out of 9 shared gene probes are among the 10 significant gene probes that can identify good from bad islets based on the qRT-PCR analysis. However, the other 5 gene probes (EST2, EST5, EHD4, MIR181A2, RNF187) in the 14 probeset were not among those included in the qRT-PCR study. Similarly, 3 (KCNMA1, RND3, CARD6) out of 4 shared gene probes in the further reduced 5 probeset were also among the same 10 significant qRT-PCR gene probes. One other shared probe had a p value of 0.08 (PMEPA1) in the qRT-PCR and the fifth (EST2) was not represented.


Transplantation of insulin-producing islet cells has been shown to be an effective treatment for severe type 1 diabetic patients. However, the effectiveness of the therapy varies greatly between islet transplantation centers [8,21]. It is widely accepted that this is predominantly due to the variability in islet preparation quality. And yet, effective and sensitive methods of gauging islet quality have been slow to develop. In this study, we hypothesized that the effectiveness of the islet graft depends both on beta cell function as well as the interaction between the graft and the host, and that these are governed by the expression of specific islet genes. Consequently, we examined the gene expression profiles of 59 human islet preparations using oligonucleotide arrays. 262 probesets, representing 199 individual genes, were identified that were differentially expressed between human islet preparations that were effective (good) or ineffective (bad) at reversing diabetes after transplantation in mice. The 262 probe classifier set predicted the ability of a specific preparation to reverse diabetes with 83% accuracy. A subset of 36 probesets had a similar predictive value, and 10 of the twenty-five genes represented in this subset were independently validated with a new set of samples by qRT-PCR.

A common theme from pathway analysis of the 262 classifier set was that a large number of significant classifiers were found to be associated with inflammation and other immune responses (Fig 3), some of which been reported to have roles in islet function and diabetes. For example, components of specific cytokine pathways are upregulated in bad islets, including tumor necrosis factor (TNF) machinery such as the TRAIL receptor TNFRSF10B, which is directly involved in T cell-induced beta cell death [22,23]. Also, both FAS and its ligand, FASL, which are associated with induction of beta cell apoptosis [24,25], are at higher levels in bad islets, suggesting that islet death-related pathways are already activated in these preparations even before transplantation. In addition to apoptosis, these pathways activate NFκB and AP-1 transcription factors, resulting in upregulation of inflammatory cytokine expression [26]. One of these, CCL2 (MCP1), is documented to promote a local proinflammatory environment associated with islet death and diabetes [1619]. Bad islets also have a higher expression of the pattern recognition receptor TLR3, which is coupled to islet dysfunction and increased cytokine expression [27]. The elevated tissue factor (F3) expression is pro-inflammatory as well and inhibits islet graft function [16,28]. Other chemokine systems are also increased, such as TGFB2 and its receptor TGFBR1 and the IL13 receptor, OSMR, but these may initiate protective signals for islet cells [2931]. Likewise, SERPINA3, also known as alpha-1-antichymotrypsin, is upregulated and in other systems is involved in wound healing [32,33]. So it appears that the pathways leading to islet dysfunction are already triggered before transplantation, but that there is also the initiation of some counteractive measures.

Conversely, a large number of genes that were preferentially upregulated in good islet preparations were associated with pancreas development and regeneration, suggesting that if repair/regeneration pathways were already initiated in damaged islets that they would be more likely to be effective after transplantation. Some of these genes include ONECUT1 (HNF6) [34,35], MNX1 (HB9) [36,37], NKX2-2 [38,39], INSM1 [40,41], NKX6-1 [38,42], FOXA2 [4345], and PTCH1 [46,47] that interact in regulatory networks (Fig 3B) guiding embryonic pancreas development and regeneration following injury. Interestingly, another molecule implicated in this process, NOTCH2, is preferentially expressed in the bad islet preparations. A possible explanation lies in the importance of NOTCH2 in expansion of the progenitor cell population by suppression of neurogenin3-dependent endocrine cell differentiation [4851].

To reduce the number of classifiers to a manageable level the 36 probeset list was subjected to logistic regression analysis with backward step-wise selection. The results indicate that the number of genes can be reduced to 14 or even 5 without loss of predictive power, though this of course must be evaluated experimentally. It is interesting to note, however, that the second highest scoring gene in this analysis was the rectifying potassium channel KCNMA1 which is upregulated in good islets and has been shown to be important for repolarization of the membrane following insulin secretion. Loss of KCNMA1 suppresses insulin secretion and increases susceptibility to oxidative stress and apoptosis [52]. Conversely, the sixth highest, SEPT9, is upregulated in bad islets and has recently been shown to be upregulated in islets of type 2 diabetics [53].

Due to the method by which the U133 Plus 2.0 GeneChips were developed, some of the best classifiers (6 of the 36 probeset list) were expressed sequence tags (ESTs) which were not mapped to coding regions. One of these, EST4, was subsequently mapped to microRNA MIR181A2, while the others appear to be in the 3’ untranslated regions of specific genes (EST1 in EGFR, EST2 in TRAPPC9, and EST3 in FOXE1). Two classifiers, EST5 and EST6, appear to be potential new genes, wojo and kyber respectively, of unknown function which were predicted by computational methods along with some expression evidence. One of these, wojo, was previously reported in a human islet cDNA screen (Melton et al, Endocrine Pancreas Consortium, unpublished).

The goal of the present study was to develop a diagnostic for assessment of the quality of cell preparations prior to use in clinical islet transplantation therapy for type 1 diabetes. After isolation, islets are typically infused into the patient within 24h-48h, and so methods of assessment must be rapid. For application of a “gene signature” of islet quality there are several methods for quantifying gene expression, and in this case we evaluated qRT-PCR analysis with a subset of 20 genes. Expression levels of half of these were significantly associated with reversal of diabetes in mice (Table 2) and together showed an 86% predictive accuracy for the outcome. Using the logistic regression model and step-wise simplification also suggest that use of as little as 5 gene probes (three of which are also represented in the 10 gene probes found significant in the qRT-PCR) could separate good from bad islet preparations with a maximum true positive predictive rate of >90% while maintaining a false positive rate of zero. Further studies will be required to determine whether this is the optimal set of classifiers for clinical application and whether qRT-PCR is the best method for utilization of this approach. For example, a limited evaluation of a bead-based RNA hybridization assay (Panomics Quantigene) with a small set of these genes provided similar discrimination between islet quality classes (data not shown). However, no matter the method of quantification it is our opinion that due to the heterogeneous nature of islet preparations that an effective diagnostic will require a set of genes and a strategy for combining the data into a meaningful metric. Logistic regression is one possible way to combine expression levels of several classifiers, such as in Fig 2 in which islet preparations that exceed a specified threshold would be considered transplantation quality.

We also investigated genes that correlated with two other standard measures of islet quality, namely glucose-responsive oxygen consumption rates (OCR) and beta cell apoptosis (BAP), in the hope of identifying one universal set of classifiers that could alone be used for preclinical islet assessment. However, it appears that each of these three parameters (reversal of diabetes in mice, OCR, and BAP) were correlated with different gene classifiers with overlap of only 9 probesets (Fig 4B). It is interesting that the genes associated with these different assays also reflect distinct biology with OCR associated with metabolism, BAP associated with various signaling pathways, and reversal of diabetes with inflammation and regeneration which require interactions within the organism. We are now of the opinion that all three of these islet assessments provide important complementary data for assessing islet function prior to transplantation.

A potentially confounding aspect of the current study is that islets contain several different cell types, each with unique gene expression profiles. This is especially true of human islet preparations, which have been shown to vary in the percentage of individual cell types by more than 300% [54]. Further complications are introduced by the effects of pancreas digestion and islet purification methods [55], as well as the islet restructuring that occurs during the unavoidable step of short-term post-isolation culture [56]. In the face of such complexity, we chose to implement an unbiased approach by profiling samples in parallel with transplantation into diabetic mice, i.e. without “islet picking”. We felt this would allow us to identify molecules that potentially affected engraftment in our islet recipients, without the bias of focusing on molecules from the insulin-producing beta cells. We are currently investigating the expression of several of the gene products in both human pancreata and the resultant islet preparations to identify the cell-type specificity of these molecules.

Another question which arises from this study is whether the gene classifiers were originally expressed in the donor organ or were expressed as a consequence of the islet isolation process. Islet isolation from human pancreata is a rigorous process involving both enzymatic and mechanical dissociation of the tissue followed by gradient separation of the cell clusters. It has been reported that this process, especially the gradient isolation, harms islets and makes them less suitable for cell therapy, although the specifics of the damage are still unknown [57]. Another recent study has shown significant changes in human islet gene expression in response to inflammatory cytokines [58]. We investigated our list of candidate molecules and found that some are present in the donor organ prior to processing and may correlate with certain islet characteristics; however, their ability to predict islet quality has yet to be determined.

An important aspect for the field of transplantation is the effective transfer of standardized diagnostic measures to other transplantation centers. This has been especially true in islet cell transplantation where the standard measures for islet assessment are known to be inadequate, but validation of new assays across centers has been difficult. We previously addressed this problem in the development of the glucose-responsive oxygen consumption assay which was validated in two centers simultaneously [15], but as far as we know this is the only new assay that has been compared in more than one center. By contrast, gene expression analysis is readily available at most research centers, and so the current approach to islet assessment may be evaluated at other centers.

In conclusion, our microarray-based analysis of 59 human islet preparations has identified a set of 262 probesets whose expression constitutes a “gene signature” of islet quality as it relates to cell therapy for diabetes. This gene set is being incorporated as part of the pre-transplant assessment of human islets for clinical transplantation therapy in our islet transplantation program. Further investigation of the role of these molecules in islet cell biology is ongoing. Moreover, the expression of the identified gene set is being determined in parallel to the OCR and BAP assays in prospective clinical trials in order to determine the relevance of each parameter and its influence on short and long term islet survival and function post-transplantation in humans.

Materials and methods

Human islet isolation and processing

Human islets were provided by the Southern California Islet Cell Resources Center (SC-ICRC) at City of Hope (Duarte, CA). The study was approved by the City of Hope Institutional Review Board and with the written informed consent from each organ donor for research use. Pancreata were digested by a modified Ricordi method [59] using Liberase-HI collagenase (Roche Molecular Biochemicals, Indianapolis, IN), then purified on a continuous Biocoll (Biochrom, Berlin, Germany) gradient in a cooled COBE 2991 Cell Processor (Gambro BCT, Lakewood, CO). Islet fractions collected from the COBE that had a purity >70% were pooled and cultured (1–2 days) in Miami Media #1 (Mediatech Inc., Herndon, VA) prior to RNA isolation and in vitro and in vivo analyses. Human islets were processed under strict GMP-compliant conditions, suitable for human clinical transplantation, using the same islet isolation protocol, facility and isolation team for each preparation. To obtain a true gene expression signature of the cell preparations used for islet cell therapy, human islet preparations were analyzed as they were received from the transplantation center, without manual selection of islets or other manipulations. Islet gene expression was analyzed using 59 individual human islet preparations. The average donor age was 44.0 ± 113.1 years (mean ± standard deviation; range 15–68 years) and 33 of the 59 (56%) pancreas donors were male. The average purity of the islet preparations was 73.6 ± 12.0% (mean ± standard deviation; range: 30–90%). Aliquots of these preparations were assessed (see below) for glucose-responsive oxygen consumption rates, beta cell apoptosis, and by transplantation into diabetic mice concomitant to RNA extraction to avoid potential bias in the gene expression introduced by differences in cell culture times.

Measurement of glucose-responsive oxygen consumption rates (OCR)

The islet flow culture system has been described in detail previously [60]. Briefly, 750 unsorted cell clusters from each islet preparation were loaded in duplicate into the inverted perfusion system and absolute levels of OCR were calculated as the flow rate times the difference between inflow and outflow oxygen tension measured by the phosphorescence lifetime of an oxygen-sensitive dye that was painted inside the perifusion chamber [15]. Inflow oxygen tension remained constant during the course of the experiment [15], and was determined at the conclusion of each experiment after inhibiting cellular respiration by the addition of antimycin A [15]. The changes in OCR in response to glucose were calculated as the difference in OCR averaged from 30 to 45 min following the change to 20 mM glucose, and the 15 min prior to the change.

Measurement of percent beta cell-apoptosis (BAP) by laser scanning cytometry

Laser scanning cytometry was performed as previously described [14]. Briefly, 500–1000 IEQ were fixed in 10% formalin, embedded in paraffin and sectioned at the City of Hope Anatomical Pathology Core or SC-ICRC facilities. Slides were immunostained for terminal deoxynucleotidyl transferase-mediated dUTP-biotin nick end labeling (TUNEL) using the ApopTag® Plus Fluorescein In Situ Apoptosis Detection Kit (Millipore/Chemicon, Temecula, CA), following manufacturer recommendations, and for insulin using guinea pig anti-human insulin antibody as primary antibody (Linco Research/Millipore, St Charles, MO) and a Cy5 conjugated secondary antibody (Jackson Immuno-Research, West Grove, PA). Slides were scanned using a iCys laser scanning cytometer (40x objective, Compucyte, Woodbridge, MA. U.S.A.) and iCys 3.2.5 software. Cells staining for insulin were defined as beta cells; cells that co-stained for insulin and TUNEL were defined as apoptotic beta cells.

Diabetes induction, islet transplantation and blood glucose monitoring

Mice were housed in specific pathogen free (SPF) conditions at the Animal Resources Center (ARC) of the Beckman Research Institute of City of Hope. NOD.SCID mice were obtained from the ARC Breeding colony at City of Hope, which were derived from breeder animals received from Jackson Laboratories (Bar Harbor, ME). The use of animals and the animal procedures were approved by the City of Hope Research Animal Care Committee. Diabetes was induced by intraperitoneal injection of streptozotocin (50 mg/kg, daily for 3 days, Sigma-Aldrich) freshly dissolved in citrate buffer. Blood samples were taken from the tail and glucose levels were measured using the One-Touch Ultra Blood Glucose Monitoring System (Lifescan Inc., Milpitas, CA). Animals were considered diabetic following two consecutive blood glucose measurements >400 mg/dL. The transplantation of 1000–2000 IEQ under the renal capsule of one kidney was performed as described previously [13,61]. Post transplant blood glucose measurements were taken two to three times per week. Islets were considered functional if the average blood glucose levels remained below 200 mg/dL, 3–4 weeks after transplantation. Islet preparations were tested in three or more animals; if the transplant successfully reversed diabetes for 3–4 weeks, the engrafted kidney was removed to ensure that glycemic reduction was dependent on the islet grafts. No reversal of diabetes was observed that was not graft-dependent.

Gene expression profiling and analysis

RNA was extracted from islet preparations using Trizol (Invitrogen). Biotinylated cRNA was prepared using the Ambion MessageAmp Biotin II kit (Ambion) and hybridized to Affymetrix Human Genome U133 Plus 2.0 GeneChips which profiles the whole known human genome representing about 47,000 transcripts. Normalized signals were generated using quantile normalization (RMAExpress[62]). Batch effects were removed using ComBat [63] and the results were used for Class Comparisons (ANOVA) and Class Predictions (BRB Array Tools; The 59 microarray datasets (data uploaded to GEO, GSE75062) were randomly assigned to two groups, the first was used to identify differentially expressed genes by Class Comparison and the second for testing the predictive power of each classifier by Class Prediction, and this randomization was repeated three times and the results pooled. Class predictions were performed using the Diagonal Linear Discriminant Analysis (DLDA) method, which is based on maximum likelihood discriminant rules that give consistently good results with our data set and others[64]. In addition to the above analysis, an independent analysis was done using the Partek Genomics Suite (Partek Inc.) to determine if the classifiers identified using BRB-Array Tools were reproducible. An ANOVA for differential expression was performed and the results compared to the genelists obtained using BRB-Array Tools. An identical approach to the BRB-Array Tools methodology was used to refine the gene signatures using 3 algorithms (DLDA, Random Forest and Linear Regression) in the Partek software. For analysis of genes associated with beta cell apoptosis and oxygen consumption rates the Class Comparison and Class Prediction each utilized the entire set of samples. Functional analysis was performed using Gene Ontology (GO) ( and Ingenuity Pathway Analysis (IPA). Receiver Operating Characteristics (ROC) analysis was done using JROCFIT ( All the microarray data for this study are available for review at the NIH GEO accession site.

QRT-PCR using OpenArray TaqMan arrays

Custom OpenArray plates were designed by comparing the available TaqMan assays from Applied Biosystems with the 36 gene classifier set obtained from the microarray analyses. The 36 Probesets represented 25 genes, and twenty of these had assays available (Table 2). Five reference genes were also chosen (HPRT1, GUSB, PPIB, ACTB and GAPDH) and post-analysis indicated that GUSB exhibited the most stable expression and so it was used for normalization of the results. Total RNA was isolated from sixteen new human islet preparations representing eight that reversed diabetes and eight that failed. The RNA samples were reverse transcribed using Superscript II (Life Technologies) and relative gene expression measured by PCR on an OpenArray NT using OpenArray Master Mix according to manufacturer instructions (Applied Biosystems). Assays were performed in duplicate for each sample on three separate days for a total of six technical replicates for each sample for each gene. The expression was quantified with the R program qpcR package[65] using the cm3 model[66]. Some of the sample wells (756 of 24192 total or 3.1%) failed amplification due to robot filling errors and the results were removed as outliers by ROUT analysis[67]. Expression values were normalized with the GUSB reference gene and technical replicates were averaged for each sample. The averages for each gene were used to assess significant differences in expression associated with islet quality.

Statistical analysis

BRB ArrayTools and Partek Genomics Suite were used for analysis of gene expression data and the internal statistical procedures were utilized with alpha and beta set to 0.05 and fold-difference set to 1.5. Analysis of qRTPCR data was done using GraphPad Prism with a p-value < 0.05 considered significant. To narrow the focus of probesets that were found to successfully classify islets into good and bad preparations, the logistic regression analyses were performed using the microarray data. Treating islet quality indicator as the dependent variable, the predicted log odds of being classified as a good islet as opposed to being a bad islet are defined as scores. The probesets were individually screened for their ability to predict the genomic profile classification, their p-values were ranked and the maximum possible subset of the classifying probes were included in a multivariate model (full model). A backward step-wise model selection procedure was then used to reduce the number of classifiers. Boxplots and ROC curves (ROCR package [68]) were used to illustrate the classification performance of the selected models. Specifically, boxplots were used to demonstrate the separation between good and bad islet groups, and the ROC curves were used to evaluate the logit model in terms of the trade-off between true positive and false positive rates. The logistic regression analyses were performed using R statistical software (version 3.1.2).


We thank Raj Puri for the helpful discussions. Funding for this work is from the National Institute of Diabetes and Digestive and Kidney Diseases, Southern California Islet Cell Resources (SC-ICR) Center, grant U42RR016607


  1. 1. Frumento D, Ben Nasr M, El Essawy B, D'Addio F, Zuccotti GV, et al. (2017) Immunotherapy for type 1 diabetes. J Endocrinol Invest 40: 803–814. pmid:28260183
  2. 2. Fiorina P, Jurewicz M, Tanaka K, Behazin N, Augello A, et al. (2007) Characterization of donor dendritic cells and enhancement of dendritic cell efflux with CC-chemokine ligand 21: a novel strategy to prolong islet allograft survival. Diabetes 56: 912–920. pmid:17287465
  3. 3. Fiorina P, Shapiro AM, Ricordi C, Secchi A (2008) The clinical impact of islet transplantation. Am J Transplant 8: 1990–1997. pmid:18828765
  4. 4. Jamiolkowski RM, Guo LY, Li YR, Shaffer SM, Naji A (2012) Islet transplantation in type I diabetes mellitus. Yale J Biol Med 85: 37–43. pmid:22461742
  5. 5. Barton FB, Rickels MR, Alejandro R, Hering BJ, Wease S, et al. (2012) Improvement in outcomes of clinical islet transplantation: 1999–2010. Diabetes Care 35: 1436–1445. pmid:22723582
  6. 6. Fiorina P, Folli F, Bertuzzi F, Maffi P, Finzi G, et al. (2003) Long-term beneficial effect of islet transplantation on diabetic macro-/microangiopathy in type 1 diabetic kidney-transplanted patients. Diabetes Care 26: 1129–1136. pmid:12663585
  7. 7. Fiorina P, Folli F, Zerbini G, Maffi P, Gremizzi C, et al. (2003) Islet transplantation is associated with improvement of renal function among uremic patients with type I diabetes mellitus and kidney transplants. J Am Soc Nephrol 14: 2150–2158. pmid:12874470
  8. 8. Shapiro AM, Ricordi C, Hering BJ, Auchincloss H, Lindblad R, et al. (2006) International trial of the Edmonton protocol for islet transplantation. N Engl J Med 355: 1318–1330. pmid:17005949
  9. 9. Piemonti L, Everly MJ, Maffi P, Scavini M, Poli F, et al. (2013) Alloantibody and autoantibody monitoring predicts islet transplantation outcome in human type 1 diabetes. Diabetes 62: 1656–1664. pmid:23274902
  10. 10. Hanson MS, Park EE, Sears ML, Greenwood KK, Danobeitia JS, et al. (2010) A simplified approach to human islet quality assessment. Transplantation 89: 1178–1188. pmid:20182409
  11. 11. Kissler HJ, Niland JC, Olack B, Ricordi C, Hering BJ, et al. (2010) Validation of methodologies for quantifying isolated human islets: an Islet Cell Resources study. Clin Transplant 24: 236–242. pmid:19719726
  12. 12. Papas KK, Suszynski TM, Colton CK (2009) Islet assessment for transplantation. Curr Opin Organ Transplant 14: 674–682. pmid:19812494
  13. 13. Gaber AO, Fraga D, Kotb M, Lo A, Sabek O, et al. (2004) Human islet graft function in NOD-SCID mice predicts clinical response in islet transplant recipients. Transplant Proc 36: 1108–1110. pmid:15194386
  14. 14. Todorov I, Nair I, Avakian-Mansoorian A, Rawson J, Omori K, et al. (2010) Quantitative assessment of beta-cell apoptosis and cell composition of isolated, undisrupted human islets by laser scanning cytometry. Transplantation 90: 836–842. pmid:20697327
  15. 15. Sweet IR, Gilbert M, Scott S, Todorov I, Jensen R, et al. (2008) Glucose-stimulated increment in oxygen consumption rate as a standardized test of human islet quality. Am J Transplant 8: 183–192. pmid:18021279
  16. 16. Bertuzzi F, Marzorati S, Maffi P, Piemonti L, Melzi R, et al. (2004) Tissue factor and CCL2/monocyte chemoattractant protein-1 released by human islets affect islet engraftment in type 1 diabetic recipients. J Clin Endocrinol Metab 89: 5724–5728. pmid:15531535
  17. 17. Igoillo-Esteve M, Marselli L, Cunha DA, Ladriere L, Ortis F, et al. (2010) Palmitate induces a pro-inflammatory response in human pancreatic islets that mimics CCL2 expression by beta cells in type 2 diabetes. Diabetologia 53: 1395–1405. pmid:20369226
  18. 18. Martin AP, Rankin S, Pitchford S, Charo IF, Furtado GC, et al. (2008) Increased expression of CCL2 in insulin-producing cells of transgenic mice promotes mobilization of myeloid cells from the bone marrow, marked insulitis, and diabetes. Diabetes 57: 3025–3033. pmid:18633103
  19. 19. Melzi R, Piemonti L, Nano R, Clissi B, Calori G, et al. (2004) Donor and isolation variables associated with human islet monocyte chemoattractant protein-1 release. Transplantation 78: 1564–1567. pmid:15599324
  20. 20. Oliveros JC (2007) VENNY. An interactive tool for comparing lists with Venn Diagrams.
  21. 21. Alejandro R, Barton FB, Hering BJ, Wease S, Collaborative Islet Transplant Registry I (2008) 2008 Update from the Collaborative Islet Transplant Registry. Transplantation 86: 1783–1788. pmid:19104422
  22. 22. Cheung SS, Metzger DL, Wang X, Huang J, Tai J, et al. (2005) Tumor necrosis factor-related apoptosis-inducing ligand and CD56 expression in patients with type 1 diabetes mellitus. Pancreas 30: 105–114. pmid:15714132
  23. 23. Ou D, Metzger DL, Wang X, Huang J, Pozzilli P, et al. (2002) TNF-related apoptosis-inducing ligand death pathway-mediated human beta-cell destruction. Diabetologia 45: 1678–1688. pmid:12488957
  24. 24. Loweth AC, Williams GT, James RF, Scarpello JH, Morgan NG (1998) Human islets of Langerhans express Fas ligand and undergo apoptosis in response to interleukin-1beta and Fas ligation. Diabetes 47: 727–732. pmid:9588443
  25. 25. Maedler K, Spinas GA, Lehmann R, Sergeev P, Weber M, et al. (2001) Glucose induces beta-cell apoptosis via upregulation of the Fas receptor in human islets. Diabetes 50: 1683–1690. pmid:11473025
  26. 26. Firestein GS, Manning AM (1999) Signal transduction and transcription factors in rheumatic disease. Arthritis Rheum 42: 609–621. pmid:10211874
  27. 27. Shibasaki S, Imagawa A, Tauriainen S, Iino M, Oikarinen M, et al. (2010) Expression of toll-like receptors in the pancreas of recent-onset fulminant type 1 diabetes. Endocr J 57: 211–219. pmid:20009359
  28. 28. Johansson H, Lukinius A, Moberg L, Lundgren T, Berne C, et al. (2005) Tissue factor produced by the endocrine cells of the islets of Langerhans is associated with a negative outcome of clinical islet transplantation. Diabetes 54: 1755–1762. pmid:15919797
  29. 29. Russell MA, Cooper AC, Dhayal S, Morgan NG (2013) Differential effects of interleukin-13 and interleukin-6 on Jak/STAT signaling and cell viability in pancreatic beta-cells. Islets 5: 95–105. pmid:23510983
  30. 30. El-Gohary Y, Tulachan S, Guo P, Welsh C, Wiersch J, et al. (2013) Smad signaling pathways regulate pancreatic endocrine development. Dev Biol 378: 83–93. pmid:23603491
  31. 31. Xiao X, Wiersch J, El-Gohary Y, Guo P, Prasadan K, et al. (2013) TGFbeta receptor signaling is essential for inflammation-induced but not beta-cell workload-induced beta-cell proliferation. Diabetes 62: 1217–1226. pmid:23248173
  32. 32. Hoffmann DC, Textoris C, Oehme F, Klaassen T, Goppelt A, et al. (2011) Pivotal role for alpha1-antichymotrypsin in skin repair. J Biol Chem 286: 28889–28901. pmid:21693707
  33. 33. Zhang B, Abreu JG, Zhou K, Chen Y, Hu Y, et al. (2010) Blocking the Wnt pathway, a unifying mechanism for an angiogenic inhibitor in the serine proteinase inhibitor family. Proc Natl Acad Sci U S A 107: 6900–6905. pmid:20351274
  34. 34. Zhang H, Ables ET, Pope CF, Washington MK, Hipkens S, et al. (2009) Multiple, temporal-specific roles for HNF6 in pancreatic endocrine and ductal differentiation. Mech Dev 126: 958–973. pmid:19766716
  35. 35. Jacquemin P, Durviaux SM, Jensen J, Godfraind C, Gradwohl G, et al. (2000) Transcription factor hepatocyte nuclear factor 6 regulates pancreatic endocrine cell differentiation and controls expression of the proendocrine gene ngn3. Mol Cell Biol 20: 4445–4454. pmid:10825208
  36. 36. Bonnefond A, Vaillant E, Philippe J, Skrobek B, Lobbens S, et al. (2013) Transcription factor gene MNX1 is a novel cause of permanent neonatal diabetes in a consanguineous family. Diabetes Metab 39: 276–280. pmid:23562494
  37. 37. Jeon J, Correa-Medina M, Ricordi C, Edlund H, Diez JA (2009) Endocrine cell clustering during human pancreas development. J Histochem Cytochem 57: 811–824. pmid:19365093
  38. 38. Lyttle BM, Li J, Krishnamurthy M, Fellows F, Wheeler MB, et al. (2008) Transcription factor expression in the developing human fetal endocrine pancreas. Diabetologia 51: 1169–1180. pmid:18491072
  39. 39. Sussel L, Kalamaras J, Hartigan-O'Connor DJ, Meneses JJ, Pedersen RA, et al. (1998) Mice lacking the homeodomain transcription factor Nkx2.2 have diabetes due to arrested differentiation of pancreatic beta cells. Development 125: 2213–2221. pmid:9584121
  40. 40. Mellitzer G, Bonne S, Luco RF, Van De Casteele M, Lenne-Samuel N, et al. (2006) IA1 is NGN3-dependent and essential for differentiation of the endocrine pancreas. EMBO J 25: 1344–1352. pmid:16511571
  41. 41. Zhang T, Wang H, Saunee NA, Breslin MB, Lan MS (2010) Insulinoma-associated antigen-1 zinc-finger transcription factor promotes pancreatic duct cell trans-differentiation. Endocrinology 151: 2030–2039. pmid:20215568
  42. 42. Schaffer AE, Taylor BL, Benthuysen JR, Liu J, Thorel F, et al. (2013) Nkx6.1 controls a gene regulatory network required for establishing and maintaining pancreatic Beta cell identity. PLoS Genet 9: e1003274. pmid:23382704
  43. 43. Ejarque M, Cervantes S, Pujadas G, Tutusaus A, Sanchez L, et al. (2013) Neurogenin3 cooperates with Foxa2 to autoactivate its own expression. J Biol Chem 288: 11705–11717. pmid:23471965
  44. 44. Lantz KA, Vatamaniuk MZ, Brestelli JE, Friedman JR, Matschinsky FM, et al. (2004) Foxa2 regulates multiple pathways of insulin secretion. J Clin Invest 114: 512–520. pmid:15314688
  45. 45. Gao N, LeLay J, Vatamaniuk MZ, Rieck S, Friedman JR, et al. (2008) Dynamic regulation of Pdx1 enhancers by Foxa1 and Foxa2 is essential for pancreas development. Genes Dev 22: 3435–3448. pmid:19141476
  46. 46. Nakayama S, Arakawa M, Uchida T, Ogihara T, Kanno R, et al. (2008) Dose-dependent requirement of patched homologue 1 in mouse pancreatic beta cell mass. Diabetologia 51: 1883–1892. pmid:18654758
  47. 47. Grieco FA, Moretti M, Sebastiani G, Galleri L, Spagnuolo I, et al. (2011) Delta-cell-specific expression of hedgehog pathway Ptch1 receptor in murine and human endocrine pancreas. Diabetes Metab Res Rev 27: 755–760. pmid:22069255
  48. 48. Apelqvist A, Li H, Sommer L, Beatus P, Anderson DJ, et al. (1999) Notch signalling controls pancreatic cell differentiation. Nature 400: 877–881. pmid:10476967
  49. 49. Lammert E, Brown J, Melton DA (2000) Notch gene expression during pancreatic organogenesis. Mech Dev 94: 199–203. pmid:10842072
  50. 50. Cras-Meneur C, Li L, Kopan R, Permutt MA (2009) Presenilins, Notch dose control the fate of pancreatic endocrine progenitors during a narrow developmental window. Genes Dev 23: 2088–2101. pmid:19723764
  51. 51. Zeggini E, Scott LJ, Saxena R, Voight BF, Marchini JL, et al. (2008) Meta-analysis of genome-wide association data and large-scale replication identifies additional susceptibility loci for type 2 diabetes. Nat Genet 40: 638–645. pmid:18372903
  52. 52. Dufer M, Neye Y, Horth K, Krippeit-Drews P, Hennige A, et al. (2011) BK channels affect glucose homeostasis and cell viability of murine pancreatic beta cells. Diabetologia 54: 423–432. pmid:20981405
  53. 53. Dayeh T, Volkov P, Salo S, Hall E, Nilsson E, et al. (2014) Genome-wide DNA methylation analysis of human pancreatic islets from type 2 diabetic and non-diabetic donors identifies candidate genes that influence insulin secretion. PLoS Genet 10: e1004160. pmid:24603685
  54. 54. Brissova M, Fowler MJ, Nicholson WE, Chu A, Hirshberg B, et al. (2005) Assessment of human pancreatic islet architecture and composition by laser scanning confocal microscopy. J Histochem Cytochem 53: 1087–1097. pmid:15923354
  55. 55. Nano R, Clissi B, Melzi R, Calori G, Maffi P, et al. (2005) Islet isolation for allotransplantation: variables associated with successful islet yield and graft function. Diabetologia 48: 906–912. pmid:15830183
  56. 56. Westermark GT, Westermark P, Nordin A, Tornelius E, Andersson A (2003) Formation of amyloid in human pancreatic islets transplanted to the liver and spleen of nude mice. Ups J Med Sci 108: 193–203. pmid:15000457
  57. 57. Ahn YB, Xu G, Marselli L, Toschi E, Sharma A, et al. (2007) Changes in gene expression in beta cells after islet isolation and transplantation using laser-capture microdissection. Diabetologia 50: 334–342. pmid:17180350
  58. 58. Eizirik DL, Sammeth M, Bouckenooghe T, Bottu G, Sisino G, et al. (2012) The human pancreatic islet transcriptome: expression of candidate genes for type 1 diabetes and the impact of pro-inflammatory cytokines. PLoS Genet 8: e1002552. pmid:22412385
  59. 59. Ricordi C, Lacy PE, Finke EH, Olack BJ, Scharp DW (1988) Automated method for isolation of human pancreatic islets. Diabetes 37: 413–420. pmid:3288530
  60. 60. Sweet IR, Gilbert M, Jensen R, Sabek O, Fraga DW, et al. (2005) Glucose stimulation of cytochrome C reduction and oxygen consumption as assessment of human islet quality. Transplantation 80: 1003–1011. pmid:16278578
  61. 61. Rush BT, Fraga DW, Kotb MY, Sabek OM, Lo A, et al. (2004) Preservation of human pancreatic islet in vivo function after 6-month culture in serum-free media. Transplantation 77: 1147–1154. pmid:15114076
  62. 62. Bolstad BM, Irizarry RA, Astrand M, Speed TP (2003) A comparison of normalization methods for high density oligonucleotide array data based on variance and bias. Bioinformatics 19: 185–193. pmid:12538238
  63. 63. Johnson WE, Li C, Rabinovic A (2007) Adjusting batch effects in microarray expression data using empirical Bayes methods. Biostatistics 8: 118–127. pmid:16632515
  64. 64. Dudoit S, Fridlyand J, Speed TP (2002) Comparison of discrimination methods for the classification of tumors using gene expression data. Journal of the American Statistical Association 97: 77–87.
  65. 65. Ritz C, Spiess AN (2008) qpcR: an R package for sigmoidal model selection in quantitative real-time polymerase chain reaction analysis. Bioinformatics 24: 1549–1551. pmid:18482995
  66. 66. Carr AC, Moore SD (2012) Robust quantification of polymerase chain reactions using global fitting. PLoS One 7: e37640. pmid:22701526
  67. 67. Motulsky HJ, Brown RE (2006) Detecting outliers when fitting data with nonlinear regression—a new method based on robust nonlinear regression and the false discovery rate. BMC Bioinformatics 7: 123. pmid:16526949
  68. 68. Sing T, Sander O, Beerenwinkel N, Lengauer T (2005) ROCR: visualizing classifier performance in R. Bioinformatics 21: 3940–3941. pmid:16096348