Convergence of the Insulin and Serotonin Programs in the Pancreatic β-Cell

All studies involving mice were approved by the University of California, San Francisco, Institutional Animal Care and Use Committee. Mice were housed on a 12-h light-dark cycle in a controlled climate. Midday of the day of vaginal plug discovery was set as embryonic day 0.5.

The Fev^−/− (22), Neurog3^−/− (23), Nkx2.2^−/− (24), Nkx6.1^−/− (11), ePet-cre (25), pdx1-cre (26), R26R (27), and Z/EG (28) mice have previously been described. All mice were maintained in the C57BL6 background.

Glucose and insulin tolerance tests were performed on fasting male mice injected intraperitoneally with glucose (2 g/kg) or insulin (0.75 units/kg). Blood glucose levels were measured from tail vein blood using the FreeStyle glucometer (Abbott Diabetes Care, Alameda, CA). Insulin was assayed using the Ultra Sensitive Rat Insulin ELISA kit (Crystal Chem, Downers Grove, IL) in serum from blood obtained from retroorbital plexus at 0 and 30 min after glucose injection.

Culture and transfection of mouse pancreatic ductal carcinoma (mPAC) cells and mouse βTC3, αTC1.9, and NIH3T3 cells were performed as previously described (29). The Min6 β-cell line was maintained in Dulbecco’s modified Eagle’s medium supplemented with 15% FBS, 100 units/mL penicillin, 100 g/mL streptomycin, and 71.5 μmol/L β-mercaptoethanol. For adenovirus experiments, mPAC cells were infected at a multiplicity of infection of 50:1 with adenoviruses encoding the basic helix-loop-helix (bHLH) transcription factors or control adenovirus encoding β-galactosidase and cultured for 2 days prior to assay for Fev mRNA.

Mouse pancreatic islets were isolated by collagenase digestion and handpicked (30). Isolated islets were incubated overnight in RPMI 1640 medium supplemented with 10% FBS, 100 units/mL penicillin, and 100 μg/mL streptomycin. Prior to secretion studies, media were replaced with the additives indicated and collected after 1 hour for insulin assays using the Ultra Sensitive Rat Insulin ELISA kit. Protein measurement by Western blot was performed as previously described (29).

Harvesting and processing of embryonic and adult mouse pancreatic tissues were performed as previously described (29). For adult brain sections, mice were perfused with saline and then 4% paraformaldehyde in PBS; brains were removed and cryoprotected in 30% sucrose in PBS and sectioned (50 μm) on a freezing microtome. Immunostaining was performed overnight at 4°C in PBS containing 1% goat serum (pancreatic tissues) or 4% goat serum, 2% BSA, and 0.1% Triton X-100 in PBS (brain tissues) (primary and secondary antisera listed in Supplementary Table 1). Slides were imaged on a Zeiss Axioskop Microscope or on a Leica TCS SL Confocal Microscope.

For RNA in situ hybridization analysis of paraffin sections (5 μm), adult mouse pancreas was processed, sectioned, and hybridized with digoxigenin-labeled riboprobes as previously described (31). Digoxigenin-labeled sense and antisense riboprobes were detected with alkaline phosphatase–coupled antidigoxigenin antibodies using BM purple (Boehringer Mannheim) as color substrate. For fluorescent in situ hybridization/ immunohistochemistry analysis, the slides were incubated with fluorescein isothiocyanate–conjugated anti-digoxigenin antibodies (Roche) followed by immunohistochemistry. The Fev sense probe gave no significant signal on mouse pancreas or brain.

Total RNA was isolated from tissues and cell lines with the RNeasy kit (Qiagen, Valencia, CA) and treated with Turbo DNase (Ambion, Austin, TX). cDNA was synthesized using SuperScript II Reverse Transcriptase (Invitrogen, Carlsbad, CA). For conventional RT-PCR (Fig. 3A), 25 ng was used per PCR reaction for 35 cycles for Fev using standard conditions. Real-time PCR assays were performed as previously described (32). Levels of assayed mRNAs were normalized to expression levels of mouse β-glucuronidase or β-actin. All oligonucleotide sequences are available on request.

Single-stranded oligonucleotides were 5′-end labeled using (γ-³²P)ATP and T4 polynucleotide kinase, annealed to excess complementary strand, and column purified. Electromobility shift assay (EMSA) buffers and electrophoresis conditions were as previously described (29), using 500 ng poly(dIdC):poly(dIdC) per 10 μL binding mix. Protein was generated in vitro with the TNT Coupled Reticulocyte Lysate System (Promega) in 50 μL total volume from 1 μg DNA; 1 μL (∼5 ng protein) of the reaction mix was then used per binding reaction. Sequences for the coding strands are shown in Fig. 5A.

βTC3 cells were transfected with the pBAT16.Pet1-FLAG vector or the control pBAT16 vector with no insert. pBAT16.Pet1-FLAG contains the mouse Fev cDNA linked in frame at the 3′end to the coding sequence for the FLAG peptide tag and inserted downstream of the cytomegalovirus immediate early promoter and the first intron of the human β-globin gene. Forty-eight hours after transfection, cells were formaldehyde fixed, washed, and lysed; DNA was purified and sheared; and Pet1-DNA complexes were precipitated as previously described (29) with anti-FLAG antibody. Approximately 10 ng immunoprecipitated DNA per reaction was assayed by PCR for specific promoter fragments.

To test for the presence of the serotonergic program in the pancreatic islet, we first measured the levels of the mRNAs encoding tryptophan hydroxylases TPH1 and TPH2, the two isoforms of the TPH enzyme that catalyze the initial and committed step in serotonin synthesis. The brain uses both isoforms, while nonneuronal tissues such as the gut express predominantly TPH1 (33). We could detect both mRNAs in the embryonic mouse pancreas, with a peak in both at embryonic day 18.5 (E18.5), shortly before birth (Fig. 1A). Both mRNAs persisted in the adult islet and could be detected at similar levels in the β-cell line βTC3 (Supplementary Table 2).

Islet expression was confirmed and localized by immunohistochemistry, which detected serotonin as early as E9.5 in a few glucagon-positive cells in the dorsal bud but not in the ventral bud (Supplementary Fig. 1A–G). Serotonin expression was detected in most of the glucagon-expressing cells at E12.5 (Supplementary Fig. 1H–O). At E14.5, most of the glucagon-expressing cells coexpressed serotonin (Supplementary Fig. 2A–D), but few of the insulin-expressing cells (Supplementary Fig. 2E–H) and none of the neurogenin3-expressing islet progenitor cells (Supplementary Fig. 2I–L) stained for serotonin. By E18.5, however, all insulin-expressing cells robustly stained for TPH and serotonin (Fig. 1B and Supplementary Fig. 3), supporting the conclusion that the peripartum peak in Tph1 mRNA levels (Fig. 1A) predominantly results from induction in β-cells. Isoform-specific antisera (34) detected both TPH1 and TPH2 by immunohistochemistry (Fig. 1C) and by Western blot (Fig. 1D) in adult mouse islets.

Embryonic pancreas, adult islets, and the islet cell lines also expressed the mRNAs encoding all of the remaining proteins required for completing the synthesis and packaging of serotonin: aromatic l-amino acid decarboxylase (dopamine decarboxylase, gene name Ddc), vesicular monoamine transporter (VMAT)2 (Slc18a2), and the serotonin reuptake transporter (Slc6a4) (Supplementary Table 2). These pancreatic cells also expressed mRNA encoding one of the serotonin autoreceptors found on serotonergic neurons, Htr1b, but not mRNA encoding a second closely related autoreceptor, Htr1a.

Cells in the pineal gland convert serotonin to melatonin via a two-step process in which the enzyme arylalkylamine N-acetyltransferase (AANAT) catalyzes the acetylation of serotonin to N-acetylserotonin, which is then converted to melatonin by acetylserotonin O-methytransferase. Because most inbred laboratory mouse strains carry a mutation in AANAT (35), we did not attempt to measure melatonin in mouse islets. Instead, we tested for AANAT mRNA in human islets but did not detect any significant amount, although TPH1 and TPH2 mRNA were readily detected (data not shown).

We next tested mice for the expression of the serotonergic E-twenty-six (ETS) transcription factor Pet1 (gene name Fev) (36), which is also expressed in some endocrine cells in the gut (37). Quantitative analysis by real-time RT-PCR revealed a peak at E14.5 in the expression of the mRNA encoding Pet1 in the embryonic pancreas and persistent robust expression in the adult islet (Fig. 2A). In situ hybridization localized Pet1 mRNA in the central regions of the embryonic pancreas at E14.5 (Fig. 2B), as well as in the serotonergic nuclei in the brainstem at E12.5 (Fig. 2C).

In the absence of antisera useful for immunohistochemical analysis of Pet1 expression, we used a line of transgenic mice in which 40 kb of genomic DNA from the region upstream of the Pet1 coding sequence controls the expression of the cre recombinase (ePet-cre mice) (25). When crossed with mice carrying the marker gene ROSA26 loxP-stop-loxP lacZ (R26R) (27), cre recombinase marks Pet-1–expressing cells and their descendents with β-galactosidase activity. We found that all insulin and most glucagon-expressing cells contained β-galactosidase in double transgenic embryos (Fig. 2D–F). At E14.5, all or almost all β-galactosidase⁺ cells also expressed the homeodomain transcription factors Isl1 (Fig. 2J), Nkx2.2 (Fig. 2K), and Pax6 (Fig. 2L) and many also expressed pancreatic-duodenal homeobox (Pdx)1 (Fig. 2G–I), HB9, and Nkx6.1 (data not shown), but no cells coexpressed β-galactosidase and the proendocrine bHLH transcription factor neurogenin3 (Fig. 2J–L). In addition, fluorescent in situ hybridization at E14.5 in the pancreas also localized Pet1 mRNA to the expression domain of Nkx2.2 (Supplementary Fig. 4A–C). In the adult pancreas, β-galactosidase was restricted to islet cells (Supplementary Fig. 5A–H). These data are consistent with the expression of Pet1 specifically in the endocrine lineage in the pancreas.

Interestingly, expression from the gene encoding the prototypical pancreatic transcription factor Pdx1 has also been reported in the brain (38). Using a strategy similar to that for tracing cells expressing the ePet-cre allele, we found that a Pdx1-cre allele (26) specifically marked serotonergic neurons in the brainstem (39) (Supplementary Fig. 5I–N) as well as pancreatic lineages.

To identify determinants of Pet1 expression in the developing pancreas, we started by testing the role of the bHLH transcription factor neurogenin3, which initiates the differentiation of islet cells from pancreatic progenitor cells but is only expressed transiently. Consistent with that transient expression, fluorescent in situ hybridization at E14.5 localized Fev mRNA in occasional cells expressing neurogenin3 (Supplementary Fig. 4D–F), while ePet1-cre–mediated recombination (which takes longer to yield detectable expression) did not (Fig. 2J–L).

Forced expression of neurogenin3 via adenovirus in the pancreatic progenitor cell line mPAC induced Fev mRNA transcription (Fig. 3A and B). Interestingly, Fev is the transcription factor gene most strongly induced in the neurogenin3-expressing progenitor cells in vivo (32). In addition, mouse embryos lacking neurogenin3 fail to express Fev mRNA in the developing pancreas (Fig. 3C). Fev mRNA similarly depends on Nkx2.2 (Fig. 3D) but not Nkx6.1 (Fig. 3E) or Lmx1b (K. Yang, Y.O., and M.S.G., unpublished data).

To test the function of Pet1 in the islet, we examined mice homozygous for a targeted disruption of the Fev gene (22). At E18.5, pancreas size and both gross and histological appearances of the pancreata in the Fev^−/− embryos were indistinguishable from their heterozygous and wild-type littermates. Staining for islet hormones did not detect any differences in the organization or size of the islets or in the numbers of the various islet cell types (Supplementary Fig. 6).

Levels of the mRNAs encoding glucagon, somatostatin, pancreatic polypeptide, and ghrelin were not significantly changed in the Fev^−/− embryos, but the mRNAs encoding the β-cell hormones insulin (Ins1 and Ins2) and islet amyloid polypeptide (Iapp) were significantly reduced (Fig. 4A). Consistent with these data, insulin content was also reduced in the pancreas of Fev^−/− embryos (Fig. 4B).

As previously described, Fev^−/− mice can reach adulthood without any apparent abnormalities except for behavioral phenotypes consistent with anxiety and depression (22). The weights of the Pet1-null animals and their wild-type littermates did not significantly differ throughout postnatal development and adulthood (Fig. 4C and data not shown). To assess glucose metabolism more closely, we performed intraperitoneal glucose tolerance tests at 12 weeks after birth. The adult Pet1-null animals cleared the glucose load significantly more slowly than their littermates (Fig. 4D). This defect was not due to decreased insulin sensitivity, since glucose levels fell as rapidly in mutant as in wild-type animals in response to injected insulin (insulin tolerance test [Fig. 4E]).

Instead, the Pet1-null animals displayed a defect in insulin secretion. At 30 min into the glucose tolerance test, serum insulin levels in the Pet1-null animals rose to approximately one-half the level in the wild-type littermate controls (Fig. 4F), despite the higher glucose levels (Fig. 4D). Islets isolated from the adult Fev^−/− animals also demonstrated a defect in insulin secretion in vitro (Fig. 4G). These defects in insulin secretion correlated with a decrease in pancreatic insulin content (Fig. 4H) in the Fev-null animals but not with any deficit in β-cell mass, which was actually increased in the null animals tested (Fig. 4I), although the increase did not reach statistical significance. Therefore, insulin gene expression and insulin production per β-cell are substantially reduced in the absence of Pet1 in both fetal and adult mice.

To explore the possibility that Pet1 directly targets the insulin genes, we examined the insulin promoters. The sequences of the rodent and human insulin gene promoters contain several potential binding sites for Pet1 with the consensus core sequence GGAA (40) (Fig. 5A). In an EMSA, in vitro produced Pet1 protein bound to a labeled oligonucleotide containing the two most proximal GGAA elements (located at −143 and –132 bp) (Fig. 5B–D), and binding with mutant probes demonstrated that Pet1 bound to the more proximal of the two sites (−132 bp) with the highest affinity (Fig. 5D). This site, previously called GG-I (41), is conserved in the human and rodent genes and lies adjacent to binding sites for Nkx2.2 and v-maf musculoaponeurotic fibrosarcoma oncogene homolog A (MafA) (5). Pet1 binding to this region of the mouse insulin promoter was also confirmed in intact β-cells by chromatin immunoprecipitation (Fig. 5E). Finally, Pet1 activated the human insulin gene promoter linked to the luciferase reporter gene when expressed in the pancreatic ductal cell line mPAC L20 (Fig. 5F).

To determine whether Pet1 targets other β-cell genes, we looked at the expression of the genes encoding key components in glucose sensing in the β-cell. Levels of the mRNAs encoding the sulfonyl receptor, ATP-sensitive K⁺ channel and glucokinase (Abcc8, Kcnj11, and Gck, respectively), were not reduced in the Fev^−/− embryos at E17.5, but Slc2a2 mRNA encoding Glut2 was reduced (Fig. 6A). Pet1 bound to both the Gck and Slc2a2 genes in β-cells as assessed by chromatin immunoprecipitation (Fig. 6C).

We also tested for changes in the expression of serotonergic genes in the pancreas of the Fev^−/− mice at E18.5. Surprisingly, we saw no reduction in the levels of any of the serotonergic mRNAs at E18.5 (Fig. 6B), despite previous evidence of their regulation by Pet1 in the brainstem (22,36); this result was confirmed by the normal serotonin immunoreactivity in the Fev^−/− pancreas at E18.5 (Supplementary Fig 7). However, in support of previous in vitro binding studies (36), Pet1 did bind to the 5′ flanking region of each of these genes in intact β-cells as assessed by chromatin immunoprecipitation (Fig. 6C). In contrast, the expression of Tph2, but not the other serotonergic genes, was reduced in Nkx2.2^−/− and Nkx6.1^−/− embryos (Fig. 6E and F).

Finally, we measured the expression of genes encoding multiple islet transcription factors, including many of those implicated in insulin gene expression. The only factor significantly reduced (albeit modestly) in expression in the pancreas at E18.5 in Fev^−/− embryos was Lmx1b (Fig. 6D), which also plays a role in serotonergic neuron development downstream of Nkx2.2 (10). Expression of Lmx1b also depended on Nkx2.2 in the pancreas (Fig. 6E).

Similarities in the developmental programs that drive the differentiation of the serotonergic neurons and pancreatic β-cells led us to examine whether the β-cells can also produce serotonin. We found that the β-cells, as well as some other islet cell types, express all of the genes required to synthesize, package, and secrete serotonin. β-Cells express both isoforms of TPH, the enzyme that catalyzes the rate-limiting step in serotonin synthesis: hydroxylation of tryptophan to 5-hydroxytryptophan. Expression of TPH1 peaks in the neonatal period in the islet. In addition, β-cells express dopamine decarboxylase, the enzyme that catalyzes the next (final) step in serotonin synthesis: decarboxylation of 5-hydroxytryptophan to 5-hydroxytryptamine (serotonin). β-Cells also express VMAT2, the transporter that loads serotonin into secretory vesicles. Interestingly, islet cells express very low levels or none of the synapse-related serotonergic genes SLC6A4 and Htr1a.

The expression of dopamine decarboxylase and the VMATs gives islet cells the ability to decarboxylate and store monoamine precursors and thus the APUD phenotype described almost 50 years ago (1,2). This capacity to take up and store serotonin has been exploited by using serotonin as a surrogate for insulin secretion (42) and by using ligands of VMAT2 for imaging β-cell mass in vivo (43). Monoamine uptake and storage are characteristics shared by many neuroendocrine cells, but since the islet cells also have TPH activity, they can specifically synthesize, store, and secrete serotonin.

The peak in Tph1 expression in islets that occurs during the perinatal period may provide one explanation for the variability in islet serotonin content seen in prior studies. We and others have also found that islet serotonin content is higher in females and during pregnancy (44–46). This variability in serotonin production by islets demonstrates a form of physiological regulation and suggests a function for islet serotonin. Given the much higher aggregate production and secretion of serotonin by the gut compared with that of the pancreatic islets, it seems unlikely that serotonin produced by the islet contributes substantially to systemic serotonin levels; but secretion by islet cells will primarily impact local concentrations and therefore could have autocrine or paracrine effects within the islet during the perinatal period and pregnancy analogous to the local effects of serotonin in the breast (47). During pregnancy, the high levels of serotonin drive β-cell proliferation (46). Since perinatal β-cells also rapidly proliferate (48,49), serotonin may have similar functions in pregnant and perinatal β-cells.

We also found that β-cells and other islet cells express the serotonergic transcription factor Pet1/Fev. In the pancreas, expression of Fev depends on the proendocrine transcription factor neurogenin3: neurogenin3 induced Fev expression in vitro, pancreatic Fev expression was lost in Neurog3^−/− embryos, and as we have previously described, Fev expression is high in the transient neurogenin3-positive endocrine precursor cells during pancreatic development (32). This neurogenin3 dependence, together with the in situ hybridization and lineage tracing data, demonstrates that Pet1 is expressed specifically in the islet lineage. In addition, as in the serotonergic neurons (10), Fev expression in the pancreas requires the downstream target of neurogenin3, Nkx2.2, but not Nkx6.1 (itself regulated by Nkx2.2 [24]). Our data from the Fev^−/− animals show, however, that none of these islet transcription factors depend on Pet1, thus placing Pet1 at the bottom of this cascade of transcription factors, as it is in the serotonergic neurons as well (10).

In serotonergic neurons, Pet1 drives the expression of the final differentiation products that characterize the mature cells, such as serotonergic genes Tph2 and Slc6a4 (22). Surprisingly, in the pancreas Pet1 was not necessary for Tph2 expression, even though we found that it bound to Tph2 and other serotonergic genes. In contrast, Tph2 expression in the pancreas did depend on Nkx2.2 and Nkx6.1, as it does in serotonergic neurons (10).

Instead, our data demonstrate that in the pancreas Pet1 regulates the expression of genes encoding key differentiated β-cell products, including the glucose transporter gene Slc2a2, Iapp, and both insulin genes. As a result, the Fev^−/− animals had defects in insulin production and secretion and impaired glucose clearance, despite compensatory increases in β-cell mass. Therefore, at the end of the transcription factor cascade, Pet1 guides the final differentiation and maturation of both serotonergic neurons and β-cells but does so by regulating overlapping but distinct sets of genes. It would be interesting to learn what role Pet1 may play in the expression of the β-cell glucose-sensing genes that are expressed in serotonergic neurons (14,15).

The developmental and functional parallels between the serotonergic neurons in the brain and the β-cells in the pancreas may have important practical implications. It must be kept in mind that pharmacological or genetic manipulations targeting the serotonergic neurons may inadvertently impact the β-cells as well—and vice versa. For example, transgenic strategies using regulatory elements from the Fev or Pdx1 genes to target either cell type will likely target both cell types. Since both β-cells and serotonergic neurons regulate glucose metabolism, this genetic overlap may confound studies of energy homeostasis in mouse models using these genes for targeting. In addition, methods developed for generating these cells from stem cells or other sources must be assessed carefully, since the overlaps in gene expression profiles may lead to the misidentification of the generated cells.

Epidemiologists have long recognized an association between the risks of type 2 diabetes and depression (50). Manifold causes likely contribute to this clinical association, but the genetic and functional similarities of the two key cell types involved in these diseases strongly suggest that some genetic or environmental insults may impair both serotonergic neurons and pancreatic β-cells and thus simultaneously increase the risk of both depression and type 2 diabetes. In addition, most drugs used to treat psychiatric disorders affect serotonergic signaling and may therefore also impact β-cells, especially during periods of high TPH activity in the islets, such as pregnancy and infancy.

Serotonin and insulin collaborate in an evolutionarily ancient partnership to regulate our response to changes in energy availability. Similarities in the function and development of the cells that produce serotonin and insulin reflect this evolutionary connection and have important implications for energy homeostasis and the pathology and treatment of diabetes.

This work was supported by NIH grants R01 DK21344 and U19 DK61245, cores from P30 DK063720, Larry L. Hillblom Foundation grants 2002/1E and 2007/1B, the Nora Eccles Treadwell Foundation, American Diabetes Association Grant 7-07-MN-21, Juvenile Diabetes Research Foundation Grant 16-2007-428, fellowship award 33-2007-187 (to H.K.), and the University of California, San Francisco (UCSF) Sandler Program in Basic Sciences.

No potential conflicts of interest relevant to this article were reported.

Y.O. designed and performed research, analyzed data, and wrote the manuscript. Y.K. designed and performed research, analyzed data, and reviewed and edited the manuscript. N.K., G.H., and H.K. performed research, analyzed data, and reviewed and edited the manuscript. J.W., N.N., and A.L. performed research. S.B.S. and R.M.G. performed research and analyzed data. L.H.T. and E.S.D. provided reagents, analyzed data, and reviewed and edited the manuscript. M.S.G. designed research, analyzed data, and wrote the manuscript.

The authors thank members of the German laboratory and John Rubenstein, Gerold Grodsky, and William Rutter of UCSF for helpful advice and criticism.

Parts of this study were presented in abstract form at the 65th Scientific Sessions of the American Diabetes Association, San Diego, California, 10--14 June 2005.