A whole organism small molecule screen identifies novel regulators of pancreatic endocrine development

Diabetes mellitus is foreseen to be the seventh leading cause of death in 2030 (according to the World Health Organization), warranting an urgent need to identify new therapeutics. During the development of type 2 diabetes, β-cells increase insulin production to compensate for insulin resistance. However, type 2 diabetes only manifests itself once β-cell failure and death occur, resulting in reduced amounts of circulating insulin (Weir and Bonner-Weir 2004, 2013).

The transcription factor pancreatic and duodenal homeobox 1 (Pdx1) is a key player in pancreas development and β-cell function/maturation. In vertebrates, Pdx1 is first expressed in the whole pancreatic endoderm, and its importance in early pancreas development has been demonstrated by the global knockout of Pdx1, which leads to pancreas agenesis and lethality (Jonsson et al., 1994; Offield et al., 1996; Hale et al., 2005). During development, an increase in PDX1 levels in endocrine progenitors is indispensable for their differentiation into β-cells (Bernardo et al., 2008). In addition, forced expression of PDX1 in endocrine progenitors leads to their conversion into β-cells at the expense of α-cells (Yang et al., 2011). In adult mice, PDX1 is specifically expressed in β-cells, and the deletion of Pdx1 from mature β-cells leads to their dedifferentiation and loss of function (Ahlgren et al., 1998; Gao et al., 2014). In addition, Pdx1 haploinsufficiency in mice leads to impaired β-cell function and apoptosis (Johnson et al., 2003). In mature β-cells, PDX1 regulates the expression of a whole network of genes important for β-cell function, including insulin and glucokinase (Ahlgren et al., 1998; Hani et al., 1999; Gao et al., 2014). Notably, and accordingly, MODY4 (maturity onset of diabetes of the young 4) is caused by mutations in PDX1, which lead to diabetes due to impaired β-cell function (Fajans et al., 2001). Together, these data point to a central role for PDX1 in pancreas and β-cell development. The possibility of regulating the network responsible for β-cell development by modulating a single transcription factor makes PDX1 an interesting target for small molecule screens.

To identify molecular pathways regulating pdx1 expression, we used the zebrafish, an animal model ideally suited for in vivo small-molecule screens (Gut et al., 2017); we developed novel reporters, and used them to screen 8256 structurally diverse compounds and subsequently investigated the top hits. Besides known modulators of pdx1 expression, we identified four interesting compounds that could be used to modulate pancreatic endoderm formation, β-cell specification and/or β-cell function. Notably, valproic acid (VPA) treatment increased pancreatic endoderm formation, while inhibition of TGFβ signaling by a pharmalogical inhibitor of Alk5 led to the α-cell to β-cell transdifferentiation. Furthermore, we tested HC toxin on human islets and in an induced pluripotent stem cell (iPSC)-derived pancreatic β-cell differentiation model, and found that it induces β-cell function, including enhanced expression of mature β-cell marker genes and enhanced insulin secretion.

In order to generate reliable transgenic lines to monitor pdx1 expression, we chose a bacterial artificial chromosome (BAC) approach over the more commonly used approach of short promoter fragments. This strategy has the clear advantage of having more, or even sometimes all, regulatory elements included in the transgene. We selected a BAC containing 100 kb upstream and 100 kb downstream of pdx1 and replaced the ATG of pdx1 with a luciferase cassette to allow a fast and quantitative readout of pdx1 expression levels (Fig. S1). An additional BAC transgenic line was made by inserting an EGFP cassette to visualize pdx1 expression at single cell resolution (Fig. S1). As expected, we observed pdx1 reporter expression in insulin (ins)-expressing β-cells at 120 hours post-fertilization (hpf) (Fig. 1A,B, Fig. S1) and co-expression with endogenous Pdx1 (Fig. 1C-C″). To assess the Tg^BAC(pdx1:luciferase2) [hereafter referred to as Tg^BAC(pdx1:Luc2)] line, we measured pdx1 promoter activity over the time period of β-cell maturation, i.e. 48-120 hpf. Coincident with the increase in β-cell maturation, we observed an increase in pdx1 promoter activity (Fig. 1D). Once β-cell maturation was achieved, pdx1 promoter activity decreased (Fig. 1D) and free glucose levels dropped (Fig. 1E) (Gut et al., 2013; Mullapudi et al., 2018).

It was recently shown that inhibiting Alk5 (also known as transforming growth factor beta receptor 1, Tgfβr1) in mammalian islets induces the expression of mature β-cell markers, including Pdx1 (Blum et al., 2014). To test whether the Tg^BAC(pdx1:Luc2) line was indeed functional and responsive to known modulators of pdx1 expression, we treated 72 hpf Tg^BAC(pdx1:Luc2) larvae with pharmacological inhibitors of Alk5 for 48 h followed by a luciferase assay. Indeed, we observed a significant induction of transgene expression (Fig. 2A). Thus, the response of our transgenic reporter to known modulators of pdx1 expression in mammals supports its use for the identification of novel compounds. The luciferase signal of the Tg^BAC(pdx1:Luc2) reporter could be modulated by the level of promoter activity, or by the number of β-cells. In order to test this latter hypothesis, we treated 72 hpf Tg^BAC(pdx1:Luc2) and Tg^BAC(pdx1:EGFP) larvae with a γ-secretase inhibitor (LY411575) to induce secondary islet formation (Parsons et al., 2009; Ninov et al., 2012). Indeed, we observed an increase in the luciferase signal concomitant with the induction of secondary islets (Fig. 2B). To visualize whether secondary islets were also marked by the Tg^BAC(pdx1:EGFP) reporter, we analyzed the treated animals by confocal microscopy. In DMSO-treated control larvae, we did not observe any Tg^BAC(pdx1:EGFP)-positive secondary islets. However, in line with the observed increase in luciferase signal, we found EGFP-positive cells along the pancreatic duct after treatment with the γ-secretase inhibitor (Fig. 2C).

Next, we designed a high-throughput platform to screen for modulators of pdx1 expression. We decided to use 72 hpf larvae because metabolic organs are formed by this stage, and these animals are still small enough to be kept in 96-well plates. Three 72 hpf Tg^BAC(pdx1:Luc2) larvae were transferred into each well of 96-well plates and treated in duplicate with small molecule libraries. We decided to use 10 µM, a concentration commonly used in whole organism screens in zebrafish (Andersson et al., 2012; Gut et al., 2013, 2017; Tsuji et al., 2014). After incubating the larvae with the compounds for 48 h, we performed a visual inspection of the 96-well plates for signs of toxicity, followed by incubation with a long half-life firefly luciferin to measure bioluminescence directly (Fig. 2D). By screening 8256 small molecules from diverse chemical libraries, we identified nine that caused the upregulation of the pdx1-driven luciferase signal (Fig. 2E, Table S1).

Interestingly, among the hit molecules, we found two known regulators of β-cell biology: SKF-86055 and 2′,3′-O-isopropylideneadenosine (Table S1). SKF-86055 inhibits Alk5 and leads to an upregulation of Pdx1 expression (Blum et al., 2014), whereas 2′,3′-O-isopropylideneadenosine is an adenosine receptor agonist, which increases β-cell replication after β-cell ablation (Andersson et al., 2012). In terms of the other hits, we did not further analyze the PI3K inhibitor due to its known function on the insulin signaling pathway; the MSK1 inhibitor was not commercially available; and the results of the EGFR inhibitor could not be confirmed when using a freshly ordered compound. Dose response curves of the four remaining compounds are shown in Fig. S3. Another top hit, chaulmoogric acid ethyl ester, has not been studied in the context of β-cell biology (Table S1). By analyzing the structure of chaulmoogric acid (Fig. S2), we hypothesized that it could be an agonist for the fatty acid receptor GPR40 (also known as FFAR1). Importantly, GPR40 is expressed by β-cells (Itoh et al., 2003), and its agonists have entered clinical trials for the treatment of diabetes (Poitout and Lin, 2013). To test this hypothesis, we measured GPR40 activation in COS-1 cells. Following the addition of chaulmoogric acid, we did indeed find a dose-dependent activation of GPR40 (Fig. S4). Overall, the identification of known modulators of β-cell mass and function further validates our screening strategy.

To analyze whether the hit compounds influence pancreatic endoderm development, we treated wild-type embryos from 5 hpf to 24 hpf followed by in situ hybridization for pdx1 expression (Fig. 3A). We tested five compounds and found an increase of the pdx1-positive pancreatic endoderm area in the embryos treated with the histone deacetylase (HDAC) inhibitors scriptaid and VPA, as well as those treated with the Alk5 inhibitor (Alk5i) (Fig. 3B-H). Next, we treated Tg(sox17:EGFP) embryos with VPA and analyzed them by confocal microscopy and could indeed confirm the increase in the pancreatic endoderm area (Fig. 3I-K). More strikingly, we found that the area of the hepatic endoderm appeared substantially decreased (Fig. 3H,I,K).

To analyze whether the hit compounds cause an increase in β-cell mass, we treated Tg(ins:H2B-GFP;ins:DsRed) larvae. This strategy allows one to identify newly specified β-cells by the expression of the fast-folding fluorophore GFP, whereas older β-cells will be labeled with GFP as well as with the slow-folding fluorophore DsRed (Hesselson et al., 2009). Even though we found no significant increase in β-cell number, there was a trend towards such an increase in larvae treated with the Alk5 inhibitor (Fig. 4A-G). More strikingly, we observed newly specified β-cells (i.e. GFP⁺DsRed⁻) after Alk5 inhibition (Fig. 4F,G, arrowheads). Owing to the close developmental relationship between α- and β-cells, we hypothesized that the new β-cells were α-cell derived. To test this hypothesis, we generated Tg^BAC(gcga:Cre) and Tg^BAC(gcga:CreERT2) lines to lineage trace α-cells. To our surprise, we found an even more intensive labeling of β-cells in Tg^BAC(gcga:Cre) and Tg^BAC(gcga:CreERT2) larvae (data not shown) than previously reported (Ye et al., 2015). Therefore, we decided to generate a Tg^BAC(arx:Cre) line to label α-cells specifically (Fig. S5). We crossed the Tg^BAC(arx:Cre) line to the Tg(insulin:loxP:mCherrySTOP:loxP:H2B-GFP) reporter line, in order to label all α-cell-derived β-cells specifically by the expression of H2B-GFP (Fig. 4H,I). By treating the double-transgenic larvae with the Alk5 inhibitor, we found a significant increase in α-cell-derived β-cells (Fig. 4H-J). In agreement with these findings, we observed an increase in bihormonal endocrine cells [Tg(gcg:EGFP)⁺ and Tg(ins:mCherry)⁺] after Alk5 inhibition (Fig. 4K-M).

β-cell differentiation can be examined in zebrafish by analyzing the formation of secondary islets following metabolic demand or Notch inhibition (Parsons et al., 2009; Ninov et al., 2013). To analyze whether the increase in pdx1 promoter-driven luciferase activity arises from increased endocrine cell differentiation, we treated the pan-endocrine reporter line Tg^BAC(neuroD:EGFP) with the different HDAC inhibitors. Blocking Notch signaling with the γ-secretase inhibitor DAPT leads to differentiation of endocrine cells along the pancreatic duct (Fig. S6, arrows) (Parsons et al., 2009; Ninov et al., 2012). However, none of the HDAC inhibitors led to an increase in secondary islet formation (Fig. S6, arrows).

Next, we evaluated whether HDAC inhibition could influence β-cell function directly. We speculated that enhanced β-cell function (i.e. insulin secretion) would lead to a change in larval glucose levels. By measuring glucose levels in whole larvae, we found that two out of the three HDAC inhibitors tested could reduce glucose levels (Fig. 5A). These results indicate that HDAC inhibition regulates β-cell function. To evaluate further whether HDAC inhibitors could reduce glucose levels in settings of hyperglycaemia, we treated larvae with the β-adrenergic agonist isoprenaline to induce a strong gluconeogenic response and elevate glucose levels (Fig. 5B) (Gut et al., 2013). Strikingly, the elevated levels of glucose by isoprenaline were completely reversed by co-administration with the HDAC inhibitors HC toxin, scriptaid and VPA (Fig. 5B). To analyze whether the HDAC inhibitors reduced glucose levels by inhibiting gluconeogenesis in the liver, we treated Tg(pck1:luciferase2) larvae. For all three HDAC inhibitors, we observed a reduction in the luciferase signal (Fig. 5C). To ensure that the glucose level lowering effect was not caused by general toxicity to the animals or their liver, we analyzed the general health of the animals as well as liver morphology and did not observe any obvious phenotypes after treatment with the HDAC inhibitors (Fig. S7).

To determine whether the glucose level lowering effect of the HDAC inhibitors depended on insulin action, we used insulin mutants. Owing to the lack of insulin, these mutant larvae exhibit elevated glucose levels. Therefore, any reduction in glucose levels in this model would indicate that the glucose-lowering effect is independent of insulin. Among the three HDAC inhibitors analyzed, none of them caused a reduction in glucose levels in the absence of insulin (Fig. 5D), indicating that HC toxin and scriptaid enhance β-cell functionality in vivo in an insulin-dependent manner. To investigate further the most promising compound, HC toxin, and analyze whether its mechanism of action depended on Pdx1 itself, we used pdx1 mutants (Kimmel et al., 2015). These mutants are characterized by elevated glucose levels in spite of the presence of β-cells (Kimmel et al., 2015). We treated pdx1 mutants with the HDAC inhibitor HC toxin and strikingly observed reduced glucose levels in heterozygous, but not homozygous, mutant animals (Fig. 5E). Together, these experiments show that the glucose level lowering effect of HC toxin depends on Pdx1 and insulin.

To determine whether the induction of Pdx1 expression by HDAC inhibition was conserved across species, we treated the murine β-cell line MIN6 with HC toxin and performed RT-qPCR for Pdx1 (Fig. 6A, Table S4) and Ins1 (Fig. 6B, Table S4) expression. We found an upregulation of Pdx1 and Ins1 expression after treatment with HC toxin (Fig. 6A,B).

To investigate further the effect of HC toxin on β-cell differentiation, we turned to a culture system to model human pancreatic β-cell development (Fig. 6C). We used a previously published differentiation protocol (Russ et al., 2015) to generate insulin-producing cells efficiently from human iPSCs, and exposed the cells to HC toxin during the last step of differentiation. Specifically, iPSCs undergoing differentiation were treated with HC toxin for 24 h on day 18, and subsequently analyzed by RT-qPCR at day 21 (Fig. 6C,D, Table S4). Strikingly, we observed an increase in the expression levels of key β-cell transcription factor genes including PDX1, NKX2.2 and NEUROD1 as well as INS after exposure to HC toxin, suggesting a conserved role for HC toxin in human β-cell differentiation (Fig. 6D, Table S4). In addition, the transcript levels of the mature human β-cell transcription factor gene MAFB was robustly induced in cells treated with HC toxin (Fig. 6D, Table S4). Taken together, these results indicate that HDAC inhibition can efficiently induce the expression of β-cell differentiation markers in vitro.

In order to test whether HC toxin treatment can also modulate the function of primary mammalian β-cells, we tested HC toxin in a glucose-stimulated insulin secretion (GSIS) assay in isolated C57BL/6J mouse islets. We isolated islets from wild-type adult mice and cultured them overnight in the presence of the HDAC inhibitors. On the next day, we performed a GSIS assay and measured released insulin by ELISA. Treatment of islets with HC toxin was associated with a higher GSIS compared with control islets (Fig. 7A). Given the importance of translating fish and murine model results into a human assay system, we next wanted to test whether the compounds identified in the zebrafish screen would work in human samples. Therefore, an islet perifusion assay was used to assess whether HC toxin treatment caused an increase in GSIS in freshly isolated human islets. Details about the human donors can be found in Table S2. In response to glucose, HC toxin-treated human islets showed significantly increased insulin release compared with DMSO-treated controls (Fig. 7B). Thus, consistent with the findings with mouse islets, treatment of human islets with HC toxin enhanced β-cell function via insulin secretion. These data show that compounds identified in the zebrafish screen can enhance β-cell function in primary human islets.

Small molecule screening in animals such as zebrafish combines the high-throughput rate of a cellular system with the complex inter-organ physiology of an intact animal (MacRae and Peterson, 2015; Gut et al., 2017). Here, we describe a novel approach for fast and reliable in vivo screening for modulators of pdx1 expression at the whole organism level. By choosing a 200 kb BAC to generate the reporter we could recapitulate pdx1 expression in cell types known to express this gene including β-cells and ductal cells (Ohlsson et al., 1993; Kritzik et al., 1999). This observation is in line with the concept that transgenic lines made with BACs recapitulate the endogenous expression more closely compared with those made with short promoter fragments. Besides the validation of our reporter with compounds known to induce Pdx1 expression in human tissue, we identified compounds with well-known roles in regulating β-cell biology, thus validating the strategy. One example is chaulmoogric acid, which we identified as a novel agonist for the fatty acid receptor GPR40. GPR40 is abundantly expressed by β-cells and it has been shown that the activation of GPR40 by agonists amplifies GSIS (Itoh et al., 2003). In addition, we identified compounds that have not been implicated in the context of pancreas and β-cell development. By using our screening strategy, we identified nine hit molecules that positively modulate the pdx1 reporter. Among the hit molecules, we identified three structurally diverse HDAC inhibitors: HC toxin (cyclic peptide HDAC inhibitor), scriptaid (hydroxamic acid HDAC inhibitor) and VPA (aliphatic acid HDAC inhibitor). Even though several HDAC inhibitors were screened in our assay, only VPA, HC toxin, scriptaid and trichostatin A (TSA) (data not shown) could reproducibly induce the pdx1 reporter. In addition, the two benzamide HDAC inhibitors MS-275 and CI-994 induced the pdx1 reporter, but only at 50 µM (data not shown).

TSA and VPA have recently been shown to enhance the pool of the proendocrine lineage (Haumaitre et al., 2009). This observation is in line with our results showing that VPA increases pancreatic endoderm size, which subsequently leads to an increase in the proendocrine lineage. In addition, work from Noël et al. (2008) showed a reduced liver size in hdac1 mutants, which is in line with our observations showing that VPA treatment decreased the area of the hepatic endoderm. It will be interesting to analyze the pancreatic endoderm in the hdac1 mutants. In zebrafish, VPA treatment has been shown to disrupt the clustering of endocrine cells (Li et al., 2016). However, this disruption could be caused by the early and long treatment period used, which included gastrulation (5-72 hpf). HC toxin and scriptaid have recently been shown to enhance glucose uptake and metabolism in mouse skeletal muscle (Tan et al., 2015; Gaur et al., 2017). However, these studies only focused on mechanisms downstream of insulin. It has also been shown that HDAC inhibition protects β-cells from cytokine-induced apoptosis (Larsen et al., 2007). As Pdx1 haploinsufficiency also leads to β-cell apoptosis (Johnson et al., 2003), it would be interesting to test whether increase of Pdx1 expression by HDAC inhibition protects β-cells from cytokine-induced apoptosis. In addition to these published observations, our data clearly show that scriptaid and HC toxin regulate glucose homeostasis by modulating β-cell function. The reduction in whole-body glucose levels could be due to an increase in insulin secretion by β-cells, increased glucose uptake by target tissues, such as muscle, and reduced gluconeogenesis in the liver. Even though VPA caused only a mild effect in our assays compared with scriptaid and HC toxin, data obtained from humans with epilepsy treated with VPA show increased postprandial insulin levels (Luef et al., 2002), confirming our observations.

HDACs alter chromatin structure and transcriptional activity by removing lysine acetylation on histones (Seto and Yoshida, 2014). However, several non-histone proteins are also regulated by lysine acetylation (Seto and Yoshida, 2014). Therefore, it is not clear whether the mechanism of action of VPA, HC toxin and scriptaid is based on epigenetic regulation by chromatin or on post-translational modifications of non-chromatin proteins. In the oncology field, it has been shown that HDAC inhibitors have anti-proliferative effects on cancer cells, and several of them are now being used in clinical trials (West and Johnstone, 2014). However, a clear mechanism of action has not yet been attained. Recently, it has been shown that protein acetylation can alter the metabolism of the cell (Zhao et al., 2010), which can be achieved by post-translational modification of non-histone cytoplasmic and mitochondrial proteins (Anderson and Hirschey, 2012; Kim et al., 2006). In fact, several HDACs have been localized in the cytoplasm and mitochondria (Bakin and Jung, 2004; Drazic et al., 2016; Herr et al., 2018). On a functional level, it has been shown that treatment with HDAC inhibitors causes a rapid increase in mitochondrial activity, arguing against regulation at the level of the chromatin (Becker et al., 2018). The observation of an increase in GSIS as early as 5 h after treatment (data not shown) indicates that this response is independent of epigenetic regulation or cell proliferation. Recent work has been reported that the deletion of HDAC3 in β-cells led to glucose intolerance (Chen et al., 2016). However, simultaneously, another group reported that the deletion of HDAC3 in β-cells improved glucose tolerance (Remsberg et al., 2017). Of interest, Remsberg et al. (2017) also showed that a PDX1 motif is enriched under HDAC3 binding peaks. Together with our work, these findings lead us to speculate that HDACs regulate Pdx1 expression to modulate β-cell function.

A new hope for the treatment of type 1 diabetes is the transplantation of stem cell-derived β-cells. Recently, several groups have reported the differentiation of β-like-cells from stem cells (Pagliuca et al., 2014; Rezania et al., 2014; Russ et al., 2015). However, the efficiency of generating β-like cells is still low and varies between groups. So far, HDAC inhibitors have not been used in β-cell differentiation protocols. Our findings indicate that the addition of HC toxin can efficiently induce the expression of differentiation markers in vitro, not only illustrating the value of our screen, but also adding another modulator to optimize protocols further for β-like-cell generation in vitro. Notably, the Alk5 inhibitor that we identified in our zebrafish screen has been shown to upregulate Pdx1 expression in human islets (Blum et al., 2014), and it is already in use in β-cell differentiation protocols (Pagliuca et al., 2014; Rezania et al., 2014; Russ et al., 2015).

Recent work has highlighted the potential of α-cells as a source for new β-cells during regeneration (Thorel et al., 2010; Courtney et al., 2013). Follow-up studies reported GABA as a possible signal driving α- to β-cell transdifferentiation (Ben-Othman et al., 2017; Li et al., 2017). However, independent studies have now shown that the use of the Gcg promoter for lineage tracing might have led to a misinterpretation of the results (van der Meulen et al., 2018; Ackermann et al., 2018). In agreement with this concern, we observed intensive β-cell labeling with our newly generated Tg^BAC(gcga:Cre) and Tg^BAC(gcga:CreERT2) lines (data not shown), which might be caused by the expression of gcga by β-cells (Tarifeno-Saldivia et al., 2017). As it has been shown that arxa is specifically expressed in mature α- but not β-cells (Tarifeno-Saldivia et al., 2017), we decided to generate a Tg^BAC(arxa:Cre) line to specifically label α-cells. Our results highlight a novel role for TGFβ signaling during α- to β-cell transdifferentiation. The potential role of TGFβ signaling in α-cells is in line with recent work profiling different endocrine cell types and showing the specific expression of the TGFβ receptor 2 gene in α-cells (Tarifeno-Saldivia et al., 2017). Further work is needed to elucidate how TGFβ signaling negatively regulates α- to β-cell transdifferentiation.

All zebrafish husbandry was performed under standard conditions in accordance with institutional (Max-Planck-Gesellschaft) and national ethical and animal welfare guidelines. Embryos were staged by hpf at 28.5°C (Kimmel et al., 1995). The following lines were used: Tg(fabp10a:EGFP)^as3 (Her, Yeh, and Wu, 2003), Tg(ins:Kaede)^jh6 (Pisharath et al., 2007), Tg(ins:dsRed)^m1018 (Shin et al., 2008), Tg(ins:Hsa.HIST1H2BJ-GFP;ins:DsRed)^s960 (Ninov et al., 2013) abbreviated Tg(ins:H2B-GFP;ins:DsRed), Tg(pck1:Luciferase)^s952 (Gut et al., 2013), Tg^BAC(neuroD:EGFP)^nl1 (Obholzer et al., 2008), pdx1^sa280 (Kimmel et al., 2015), insulin^bns102 (Mullapudi et al., 2018), Tg(insulin:loxP:mCherrySTOP:loxP:H2B-GFP; cryaa:Cerulean)^s934 (Hesselson et al., 2011), Tg(sox17:GFP)^s870 (Sakaguchi et al., 2006), Tg(Tp1bglob:H2BmCherry)^S939 (Ninov et al., 2012), Tg(ubb:loxP:CFP:loxP:H2B-mCherry)^jh63 (Wang et al., 2015), Tg^BAC(arxa:Cre)^bns250 (this study), Tg^BAC(pdx1:EGFP)^bns13 (this study), Tg^BAC(pdx1:luciferase2)^bns17 (this study) and Tg(ins:loxP-Eco.NfsB-mCherry-loxP-luciferase)^bns152 abbreviated as Tg(ins:mCherry) (this study).

Islets were isolated from pancreata procured within the Nordic Network for Clinical Islet Transplantation as previously described (Goto et al., 2004). Consent for organ donation (for clinical transplantation and for use in research) was obtained via an online database (https://www.socialstyrelsen.se/ansok-och-anmal/donationsregistret/) or verbally from the deceased's next of kin by the attending physician. The consent was documented in the medical records of the deceased in accordance with Swedish law and as approved by the Regional Ethics Committee (Dnr 2009/371/2). The experiments using human islets in this study were approved by the Regional Ethics Committee in Uppsala (Dnr 2015/444).

Human iPSC lines (iXM001 and BIH004) were maintained on Geltrex-coated (Invitrogen) plates in home-made E8 media, as reported by Chen et al. (2011), under hypoxic conditions. The medium was changed daily and cells were passaged every ∼3 days as cell clumps or single cells using 0.5 mM EDTA (Invitrogen) or Accutase (Invitrogen), respectively. Medium was supplemented with 10 μM Rho-associated protein kinase (ROCK) inhibitor Y-27632 (Sigma) when iPSCs were thawed or passaged as single cells.

Differentiation was carried out following a 21-day protocol described by Russ et al. (2015). Briefly, iPSCs were dissociated using Accutase and seeded at a density of 5.5×10⁶ cells per well in ultra-low attachment 6-well plates (Thermo Fisher Scientific) in 5.5 ml E8-home medium supplemented with 10 μM ROCK inhibitor, 10 ng/ml activin A (R&D Systems) and 10 ng/ml heregulin-b1 (Peprotech). Plates were placed on an orbital shaker at 100 rpm to induce sphere formation at 37°C in a humidified atmosphere containing 5% CO₂.

To induce definitive endoderm differentiation, cell clusters were collected after 36 h in a 50-ml Falcon tube, washed with PBS and re-suspended in d1 media [RPMI (Invitrogen) containing 0.2% FBS, 1:5.000 ITS (Invitrogen), 100 ng/ml activin A and 50 ng/ml WNT3a (R&D Systems)]. Clusters from two wells were combined into one well and distributed into low-attachment plates in 5.5 ml d1 media. Subsequently, cell clusters were differentiated into β-like cells by exposure to the appropriate media as previously published (Russ et al., 2015).

To generate the pdx1 and arx BAC constructs, we used the BAC clones DKEY-244H6 containing 200 kb of the pdx1 locus and CH73-369P3 containing 200 kb of the arx locus. All recombineering steps were performed as described by Bussmann and Schulte-Merker (2011) with the modification that we inserted a myl7:tagRFP-iTol2 cassette into the Tg^BAC(pdx1:luciferase)^bns17 and a cryaa:CFP-iTol2 cassette into the arxa^BAC:Cre construct to identify the transgenic animals. In addition, we removed the Kanamycin cassette with a flipase for the arxa^BAC:Cre construct. The following homology arms were used to generate the targeting PCR products of the EGFP_Kan, Luc2_Kan and Cre_Kan cassettes: pdx1-HA1, 5′-ggcgctggctcatgtgctcgtgtacggcacggtttccccggtctatggca-3′; pdx1-HA2, 5′-aatcgggaagagcattactatccgcctaaccacctgtacaaggactcttg-3′; arxa-HA1, 5′-tgcgagagagagtaacagtcacccactcgagcacgactgaggacgataca-3′; arxa-HA2, 5′-ttgcttttacattcgctccgatcgcggctatcgtcgtcgtactgactgct-3′.

The Tg^BAC(pdx1:EGFP)^bns13 and Tg^BAC(pdx1:luciferase)^bns17 lines exhibit expression in cell types known to express pdx1 including β-cells and ductal cells (Ohlsson et al., 1993; Kritzik et al., 1999). We hypothesize that due to the fast transitory stage of the pancreatic endoderm in zebrafish, the Tg^BAC(pdx1:EGFP)^bns13 and Tg^BAC(pdx1:luciferase)^bns17 lines do not clearly label the pancreatic endoderm. Another explanation would be that the regulatory elements driving pdx1 expression in the pancreatic endoderm might not be present in the DKEY-244H6. To generate the Tg(ins:loxP-Eco.NfsB-mCherry-loxP-luciferase)^bns152 line, abbreviated as Tg(ins:mCherry), the loxP Eco.NfsB-mCherry loxP cassette was amplified by PCR and cloned downstream of the insulin promoter together with a 5′ beta-globin intron. In a second cloning reaction, luciferase was amplified by PCR and cloned downstream of Eco.NfsB-mCherry loxP. Several founder fish were identified and analyzed. All founder fish exhibited similar patterns of EGFP, luciferase, Cre or mCherry expression, ruling out effects related to the site of integration.

Single in situ hybridizations were performed as described (Thisse and Thisse, 2008; Helker et al., 2013).

The following probes were synthesized: insulin (Milewski et al., 1998); pdx1 was amplified from cDNA of 24 hpf embryos using the primers 5′-caggtagagcagaggtcctga-3′ (pdx1-forward) and 5′-tcacatcactttaatgtttgtggtaa-3′ (pdx1-reverse). The T7 promoter was added to the 5′ end of the reverse primer in a second round of amplification.

Homozygous Tg^BAC(pdx1:luciferase)^bns17 animals were outcrossed to AB wild types to collect large numbers of hemizygous embryos for experiments. At 72 hpf, healthy larvae were selected and washed with egg water before distributing three larvae per well in 200 μl egg water in 96-well plates. Each compound was tested in duplicate. Drugs were used at a concentration of 10 μM in 1% DMSO. After 48 h of treatment, the samples were incubated in Steady Glo (Promega) as described previously (Gut et al., 2013). The bioluminescence signal was analyzed by FLUOstar Omega (BMG LABTECH). Each well was normalized to the average of the DMSO wells in each plate.

An upregulation of 1.5-fold observed in both replicates was assigned as the hit compound threshold. The hits were further tested in validation experiments and by glucose level measurements. The compounds were not very stable and induction of the Tg^BAC(pdx1:luciferase) reporter at low concentrations was only achieved with freshly ordered compounds. The following chemical libraries were used: LOPAC (Sigma-Aldrich), MicroSource Spectrum Collection, Prestwick Chemical Library, PKIS (Published Kinase Inhibitor Set), Selleckchem Kinase Inhibitory Library, Epigenetic Screening Library, Cayman and TimTec Natural Compound Library. Data analysis was performed using a custom script in the statistical language R.

After drug treatment, three replicates of five larvae per condition were homogenized and free glucose levels were determined using a glucose assay kit (BioVision) as described previously (Gut et al., 2013).

Total RNA from control and drug-treated MIN6 cells was extracted using an RNeasy Mini Kit (Qiagen). Reverse transcription polymerase chain reaction (RT-PCR) was performed using a SuperScript III First Strand Synthesis System (Invitrogen) according to manufacturer's instructions. RT-qPCR was carried out to quantify gene expression levels on a CFX connect Real-time System (Bio-Rad) with the following Taqman probes: Pdx1 Mm00435565_m1, Ins1 Mm01950294_s1. Each sample was normalized to the housekeeping probe Gapdh Mm99999915_g1.

Total RNA of β-like cells was isolated using a High Pure RNA Isolation Kit (Roche) or Trizol (Invitrogen). First-strand reverse transcription was performed with 1.5 μg RNA using the Transcriptor First Strand cDNA Synthesis Kit (Roche). RT-qPCR was carried out in triplicate using the SYBR Green PCR Master Mix (Thermo Fisher) on a Bio-Rad LightCycler. Results were normalized to GAPDH transcripts. Primers used for RT-qPCR are listed in Table S3. All reactions were performed with annealing at 60°C for 40 cycles. For undetectable transcripts, the cycle number was set to 40 for comparisons.

Zebrafish larvae were mounted in 1% low melt agarose. Egg water and agarose were supplemented with 19.2 mg/l Tricaine. All fluorescent images were acquired using an upright Zeiss LSM 780, 800 or 880 confocal microscope. Maximum projection images were analyzed and generated using Imaris (Bitplane).

Zebrafish larvae were immunostained according to Lancman et al. (2013). The following primary antibodies were used: Nkx6.1 (1:100, Developmental Studies Hybridoma Bank) and Pdx1 (1:100; a gift from Christopher V. Wright, Vanderbilt University, Nashville, TN, USA). The Pdx1 has been used by several other groups (Lu et al. 2016; Ye et al. 2015). However, our publication is the first time that the antibody is validated by the use of the Tg(pdx1:EGFP) line.

Pancreatic islets were isolated from 16- to 20-week-old C57BL/6J mice following LiberaseTL (Roche) infusion via the common bile duct to disrupt the pancreatic exocrine tissue and filtration to recover intact islets. Islets were hand-picked and cultured overnight in RPMI1640 supplemented with 10% FBS, 100 units/ml penicillin and 100 µg/ml streptomycin. Islets were treated with 10 µM of the hit compounds for 12 h in RPMI1640 containing 0.2% BSA, 100 units/ml penicillin and 100 µg/ml streptomycin. Glucose-stimulated insulin secretion was conducted in KRB buffer containing 4.6 mM KCl, 2.5 mM CaCl₂, 1.2 mM MgSO₄, 1.2 mM KH₂PO₄, 17.7 mM NaHCO₃, 10 mM HEPES, 117 mM NaCl and 0.2% BSA. Ten size-matched islets were pre-incubated for 1 h in KRB containing 3 mM glucose prior to collection of the supernatant from incubations with fresh KRB containing 3 mM and 15 mM glucose. Insulin secretion was assessed with a Mouse Insulin ELISA kit (Alpco) following manufacturer's protocol.

COS-1 cells were seeded in 96-well plates with white walls and transparent base and transfected with plasmids containing cDNA encoding a calcium-sensitive bioluminescent fusion protein consisting of aequorin and GFP (Baubet et al., 2000) as well as the indicated receptors using Lipofectamine 2000 (Life Technologies) following manufacturer's instructions. Forty-eight hours later, cells were loaded with 5 µM coelenterazine (Promega) in HBSS containing 1.8 mM calcium and 10 mM glucose for 2 h at 37°C. Measurements were performed using a luminometric plate reader (Flexstation 3, Molecular Devices). The area under each calcium transient was calculated using SoftMaxPro software and expressed as area under the curve (AUC).

Thirty human islets per well were handpicked under a light microscope into 6-well plates containing 2 ml of CMRL-1066 (ICN Biomedicals) supplemented with 10 mM HEPES, 2 mM L-glutamine, 50 µg/ml gentamicin, 20 µg/ml ciprofloxacin, 10 mM nicotinamide, and 10% heat-inactivated human serum. The islets were cultured for 24 h at 37°C with the addition of 10 µM HC toxin (or DMSO-containing vehicle only) before assessment of glucose-stimulated insulin secretion (GSIS) in a dynamic perifusion system (PERI-4.2, Biorep Technologies). In brief, the islets were transferred to a perifusion chamber and perifused with low (1.6 mM) and high (20 mM) glucose Krebs solution at a flow rate of 100 µl/min at 37°C and collection of the perifusate in 1 min fractions. After a 34 min stabilization period at low glucose, the islets were challenged with high glucose for 15 min and then perifused again with low glucose for 15 min. Insulin was measured in the perifusate fractions with ELISA (Mercodia, Uppsala, Sweden) according to manufacturer's instructions.

The average β-cell number counted in the islets of control DMSO-treated larvae was set at 100% and the relative changes were calculated. The pdx1-positive area was quantified using the area measurement tool from ImageJ.

Standard error of the mean and P-values from a two-tailed t-test were calculated using Microsoft Excel. Statistical significance was indicated as: *P≤0.05, **P≤0.01, ***P≤0.001.

We thank Dirk Meyer and Robin Kimmel for kindly providing the pdx1^sa280 fish line, Sonja Sievers for advice regarding the chemical libraries, Marianne Ploch and Khrievono Kikhi for technical assistance, Michelle Collins, Zacharias Kontarakis and Albert Wang for comments on the manuscript, and members of the Stainier lab for sharing reagents and discussions.