γ-Aminobutyric Acid (GABA) Is an Autocrine Excitatory Transmitter in Human Pancreatic β-Cells

Human pancreases were obtained with ethical approval and clinical consent from nondiabetic donors. Islets were isolated in the Diabetes Research and Wellness Foundation Human Islet Isolation Facility by collagenase digestion (Serva, Heidelberg, Germany), essentially as reported previously (14,15). For hormone release measurements, the islets were cultured overnight in Connaught Medical Research Laboratories (CMRL) medium containing 5 mmol/l glucose. The electrophysiological experiments were performed on single cells or cell clusters obtained by dissociation of islets in Ca²⁺-free buffer (16). The resulting cell suspension was then plated onto plastic Petri dishes and cultured in RPMI-1640 containing 10 mmol/l glucose and 2 mmol/l l-glutamine. For biophysical detection of GABA release (Figs. 3,FIG. 4–5), cells were infected with recombinant adenoviruses encoding the GABA_AR α₁ and β₁ subunits 24–48 h before the experiments (17).

Patch pipettes were pulled from borosilicate glass and heat-polished (tip resistance 4–8 MΩ). Petri dishes were mounted onto an Axiovert 10 microscope (Zeiss, Jena, Germany) positioned on a vibration isolation table with a Faraday cage (TMC, Peabody, MA). Experiments were performed in the standard or perforated-patch whole-cell configuration using an EPC9 amplifier and Pulse software (HEKA, Lambrecht, Germany). All electrophysiological measurements were conducted at 32–33°C and cells were continuously superfused with extracellular medium. For rapid application of GABA (Figs. 1 and 7), a Nanoliter 2000 Oocyte Injector (WPI, Stevenage, U.K.) was used.

The extracellular solution for measuring glucose- or tolbutamide-induced GABA release (Fig. 4) and membrane potential (Fig. 7) contained (in mmol/l) 138 NaCl, 5.6 KCl, 2.6 CaCl₂, 1.2 MgCl₂, 5 HEPES (pH 7.4, NaOH), and glucose at the indicated concentration. For detecting expression of endogenous GABA-activated currents (Fig. 1) and GABA-release elicited by Ca²⁺ infusion or voltage-clamp depolarizations in infected cells (Figs. 3 and 5), TEACl (20 mmol/l) was added and NaCl correspondingly reduced. In Fig. 1, the intracellular solution was composed of (in mmol/l) 120 CsCl, 1 MgCl₂, 10 EGTA, 1 CaCl₂, 10 HEPES, and 3 MgATP (pH 7.2, CsOH). In Fig. 4, CsCl was replaced equimolarly with KCl. The Ca²⁺ infusion experiments (Figs. 3 A and 5 A) were performed with pipette solution consisting of (in mmol/l) 110 CsCl, 10 KCl, 10 NaCl, 1 MgCl₂, 3 MgATP, 0.1 cAMP, 5 HEPES, 9 CaCl₂, and 10 EGTA (pH 7.15 with CsOH; free Ca²⁺ ∼2 μmol/l). Depolarization-evoked GABA release was recorded with intracellular medium containing (in mmol/l) 125 Cs-glutamate, 10 CsCl, 10 NaCl, 1 MgCl₂, 5 HEPES, 50 μmol/l EGTA, 3 MgATP, and 0.1 cAMP (pH 7.15 with CsOH). The membrane potential measurements were performed with a pipette solution composed of (in mmol/l) 76 K₂SO₄, 10 KCl, 10 NaCl, 1 MgCl₂, 5 HEPES, and 10–50 μg/ml gramicidin (pH adjusted to 7.35 with CsOH, osmolarity adjusted to ∼300 mOsm with sucrose). Biocytin (0.5 mg/ml) was sometimes added to the pipette solution to facilitate subsequent immunocytochemical identification.

Cells were preloaded with serotonin by addition of 0.5 mmol/l 5-hydroxytryptophan and 0.5 mmol/l serotonin to the culture medium >6 h before the experiment. Serotonin release was detected using a carbon fiber electrode (ProCFE, Dagan Corporation, Minneapolis, MN) connected to the second head-stage of the EPC9/3 amplifier. The electrode was held at 650 mV and positioned ∼1 μm from the cell. Exocytosis was stimulated by infusion of cells with solution containing 2 μmol/l free Ca²⁺ via the patch electrode (see above), and whole-cell GABA-activated currents were measured in parallel from the same cell.

De-paraffinized human pancreatic tissue sections (4 μm) were heated to 95°C in 10 mmol/l Tris + 1 mmol/l EDTA (pH 6) for 40 min, followed by incubation at 4°C for 10 min in the same buffer. Sections were then incubated with antibodies directed against GABA_AR subunits α_1–6 (1:50, H-300; Santa Cruz Biotechnology, Santa Cruz, CA), β_2/3 (1:100, MAB341, Millipore, Hampshire, U.K.) or γ₂ (1:100, LS-C14, Life span, Seattle, WA) dissolved in REAL antibody diluent (Dako, Cambridge, U.K.) at 4°C overnight. Endogenous peroxidases were blocked using methanol containing 1% H₂O₂ (15 min). A peroxidase-conjugated secondary antibody (Dako) was added for 30 min, followed by Alexa 488–labeled tyramide (Invitrogen) for 20 min. Co-staining for insulin, glucagon, somatostatin, or pancreatic polypeptide and visualization of fluorescence were performed as previously described (18).

Identification of the cell type after patch-clamp experiments was performed by immunodetection of insulin, glucagon, and somatostatin as described previously (19,20) (see the online appendix available at http://diabetes.diabetesjournals.org/cgi/content/full/db09-0797/DC1).

Batches of 10–20 islets (in triplicates) were preincubated in 1 ml Krebs-Ringer buffer containing 2 mg/ml BSA and supplemented with 1 mmol/l glucose for 1 h followed by a 1-h test incubation in 1 ml Krebs-Ringer buffer supplemented as indicated. The hormone content of the supernatant was measured by radioimmunoassay (insulin, glucagon: Millipore, Watford, U.K.; somatostatin: Euro-Diagnostica, Malmö, Sweden).

Gene expression profiling of GABA_AR subunits was performed by RT-qPCR on human islet total RNA as described previously (19) (see the online appendix).

Isolated human islets were fixed in 2.5% glutaraldehyde in phosphate buffer, postfixed in 1% OsO₄, dehydrated, and embedded in Spurr's resin. Ultrathin sections cut onto nickel grids were immunolabeled for GABA using a rabbit polyclonal antibody (1:1,000 dilution, Sigma) and protein A gold particles (15 nm, Biocell, Cardiff, U.K.). Sections were viewed with a Joel 1010 microscope (accelerating voltage 80 kV).

All data are expressed as means ± SEM. Statistical significances were calculated using Student's t test. GABA-induced transient inward currents (TICs) and amperometric events were analyzed using MiniAnalysis software (Synaptosoft, Decatur, GA).

Functional expression of endogenous GABA_AR Cl⁻ channels in human islet cells was investigated in patch-clamp experiments. Figure 1,A shows a representative recording in which puffer application of GABA to a β-cell triggered a rapidly activating and desensitizing inward current that was sensitive to the GABA_AR antagonist SR-95531. GABA-activated currents were found in 26 of 50 β-cells (52%), and the current amplitude in receptor-positive cells averaged 64 ± 20 pA (current density 9.4 ± 2.1 pA/pF). The GABA-evoked current was decreased by 78 ± 2% in the presence of SR-95531 (from 32 ± 5 to 7 ± 2 pA, P < 0.01, n = 8). GABA application also elicited inward currents in 31 out of 34 δ-cells (91%), but the responses were typically much larger than in β-cells (Fig. 1,B). On average, the GABA-evoked currents had an amplitude of 668 ± 188 pA (148 ± 42 pA/pF, n = 31), which was reduced by 75 ± 6% (from 570 ± 330 to 33 ± 8 pA, P < 0.01, n = 10) in the presence of SR-95531. By contrast, GABA application triggered inward currents in only 3 out of 48 α-cells (6%; Fig. 1,C), with the current amplitude in these cells averaging 7 ± 3 pA (current density 2 pA/pF). Insulin has been reported to translocate GABA_AR to the cell surface in α-cells (8). In α-cells that had been preincubated with insulin (0.5 μmol/l for 1 h in the culture medium), application of GABA in the continued presence of insulin (0.1 μmol/l) elicited inward currents in 3 out of 19 cells tested (16%), but the current amplitude was not increased and averaged 4.3 ± 0.6 pA/pF in these three cells. As shown in Fig. 1 C, SR95531 reduced the GABA-activated current in α-cells by 70% (n = 2). Of the 64 cells analyzed in which GABA evoked a measurable current, only one cell did not contain insulin, glucagon, or somatostatin, and this cell was not positive for pancreatic polypeptide (not shown).

Expression of GABA_AR subunits in human islets was investigated by quantitative RT-PCR. Among the α- and β-subunits, α₂ and β₃ predominated. Fairly high levels of the γ₂ and π were also observed. In addition, low levels of α₁, α₃, α₄, β₁, β₂, γ₁, and θ were detected (Fig. 2 A).

Expression of GABA_AR subunits in human islets was confirmed by immunohistochemistry. An antibody against the β_2/3 subunit labeled ∼50% of the β-cells in pancreatic tissue sections (Fig. 2,B); no staining of α-cells, δ-cells, or PP-cells was observed (supplementary Fig. 1, available in the online appendix). An α_1–6–specific antibody stained 22% of the β-cells (Fig. 2,C), 94% of the δ-cells, and 14% of the α-cells (supplementary Fig. 1). Finally, an antibody directed against the γ₂ subunit labeled β-cells (Fig. 2 D) as well as α- and δ-cells (supplementary Fig. 1). No specific labeling of human pancreatic sections was observed with an antibody against β₁ subunits (not shown).

To determine if GABA is released from human β-cells, a patch-clamp–based assay was used (17). Briefly, GABA_ARs were overexpressed in isolated β-cells using adenoviral vectors encoding the GABA_AR α₁ and β₁ subunits, respectively. GABA released upon fusion of GABA-containing secretory vesicles activates these receptors in the same cell, giving rise to transient inward currents (TICs). Figure 3 A shows a recording from a cell held at −70 mV, in which exocytosis was stimulated by inclusion of 2 μmol/l free Ca²⁺ in the intracellular solution. The trace displays numerous TICs that were sensitive to the GABA_AR blockers SR-95531 (10 μmol/l; n = 4) and bicuculline (100 μmol/l; n = 2; not shown), demonstrating that they reflect activation of GABA_AR. The TICs had an average rise time (10–90%) of 14 ± 1 ms and a half-width of 40 ± 2 ms (n = 274 events from four cells). These values are comparable to those reported in rat β-cells (17).

Quantal release of GABA could be elicited by voltage-clamp depolarizations from −70 to 0 mV (Fig. 3 B). Depolarization-evoked transient currents were outward because the experiments were conducted at low [Cl⁻]_i (22 mmol/l). Thus, the Cl⁻ equilibrium potential (E_Cl) is approximately −50 mV, resulting in a net inward driving force at 0 mV for Cl⁻ ions.

GABA release was also observed in δ-cells stimulated by Ca²⁺ infusion or voltage-clamp depolarizations (supplementary Fig. 2A). As δ-cells express high levels of endogenous GABA_AR, GABA-induced TICs were sometimes (three cells) seen even in noninfected cells (supplementary Fig. 2B).

Figure 3,B demonstrated that GABA is released from β-cells in response to depolarization-induced Ca²⁺ influx via voltage-gated Ca²⁺ channels. We went on to investigate whether glucose, via membrane depolarization, can evoke GABA release. These experiments were performed on clusters of islet cells overexpressing α₁/β₁ GABA_AR. The membrane potential of the patch-clamped cell was held at −70 mV and its cytoplasm infused with a buffer containing 10 mmol/l EGTA. These steps prevent exocytosis in the patch-clamped cell, and it will therefore serve as a sensor (“sniffer-cell”) for GABA released from neighboring cells within the cluster. TICs evoked by GABA release were rarely observed at 1 mmol/l extracellular glucose, but increasing the glucose concentration to 6 or 20 mmol/l increased the frequency of TICs after a delay of 1–2 min (Fig. 4,A–C). The glucose-induced TICs were sensitive to SR-95531 (Fig. 4,B), confirming that they reflect GABA release and activation of GABA_AR. When they occurred, the TICs tended to occur in bursts (Fig. 4 B, ii).

GABA release was likewise stimulated by addition of the ATP-sensitive K⁺ (K_ATP) channel blocker tolbutamide (100 μmol/l) to the extracellular solution (Fig. 4,D; n = 4) and inhibited by the K_ATP channel opener diazoxide (n = 4, supplementary Fig. 3). The change in holding current in Fig. 4 D probably reflects closure of K_ATP channels that remain active in the patch-clamped cells despite the presence of 3 mmol/l ATP in the intracellular solution. By contrast, glucose stimulation did not consistently affect the holding current.

In cell clusters overexpressing α₁/β₁ GABA_AR, addition of SR-95531 reduced both the holding current and current noise even at low (1 mmol/l) extracellular glucose, in the absence of observable TICs (Fig. 4 E). In a series of seven experiments, SR-95531 decreased the holding current from −47 ± 15 to −36 ± 15 pA (P < 0.05) and the current variance from 9.8 ± 2.7 × 10⁻²⁴ A² to 4.1 ± 1.8 × 10⁻²⁴ A² (P < 0.05). From the excess noise (sensitive to SR-95531), we estimate the size of the unitary event as ∼0.5 pA, corresponding to a single-channel conductance of 7 pS. This is in reasonable agreement with the 10–15 pS reported for recombinant α₁β₁ GABA_AR Cl⁻ channels (21).

GABA and GAD65 have been reported to be associated with synaptic-like microvesicles in rodent β-cells (22). However, we have recently presented evidence for release of GABA from insulin-containing large dense-core vesicles (LDCV) in rat β-cells (23). To determine whether vesicular GABA secretion from human β-cells reflects exocytosis of insulin-containing granules, we combined the patch-clamp–based assay for GABA release with amperometric detection of serotonin, assumed to accumulate in LDCVs (24,,–27). As shown in Fig. 5 A, the transient currents induced by vesicular release of GABA (top) and serotonin (bottom) occurred simultaneously in most cases. In this experiment, 90 out of 165 GABAergic TICs were accompanied by a simultaneous amperometric spike. In a series of four experiments, co-release of serotonin was detected for 44 ± 9% of the GABA-release events. This suggests that GABA and serotonin are released by exocytosis of the same vesicles; the correlation is <100% because amperometry only detects exocytotic events occurring in the vicinity of the carbon fiber (usually <50%), whereas GABA release is detectable over the entire cell surface (23). Close inspection of the records indicated that GABA may also be released during the “pedestals” that precede full fusion and during kiss-and-run exocytosis (supplementary Fig. 4).

Not all amperometric events were accompanied by simultaneous release of GABA (Fig. 5,A). Indeed, only 13 ± 3% (n = 4) of the amperometric events were associated with GABA-evoked TICs, suggesting that GABA is only present in a subpopulation of insulin granules. This was verified by immunogold electron microscopy using a GABA-specific antibody. Some insulin granules were strongly labeled with gold particles, whereas other granules were not stained at all (Fig. 5 B). This finding is in agreement with previous results from rat β-cells (23). Immunogold electron micrographs confirmed that non–β-cells (including δ-cells) contain GABA at levels 50–70% of those detected in β-cells (supplementary Fig. 5).

In intact human islets, glucose at a concentration as low as 3 mmol/l stimulated insulin secretion above that seen at 1 mmol/l glucose. At higher glucose concentrations, there was a concentration-dependent further stimulation of insulin release (Fig. 6). The GABA_AR antagonist SR-95531 had no effect on basal insulin release but inhibited insulin release at 6 mmol/l glucose. No (statistically) significant inhibition of insulin secretion was observed at 3, 10, or 20 mmol/l glucose.

Somatostatin secretion was stimulated by elevation of the glucose concentration to ≥3 mmol/l (supplementary Fig. 6A). Glucagon secretion was suppressed when the glucose level in the medium was raised from 1 to ≥3 mmol/l (supplementary Fig. 6B). Inhibition was maximal at 6 mmol/l glucose and gradually reduced at higher concentrations (28). SR95531 had complex and glucose-dependent effects on glucagon and somatostatin secretion.

The effect of activation of endogenous GABA_ARs on the membrane potential of islet cells was studied in noninfected cells using the perforated-patch whole-cell configuration. To maintain normal [Cl⁻]_i, the Cl⁻-impermeable antibiotic gramicidin was used as the perforating agent (29). Figure 7,A shows a β-cell exposed to 6 mmol/l glucose that was not electrically active. We point out that insulin secretion at 6 mmol/l is only ∼40% of that evoked by 20 mmol/l glucose. The finding that some β-cells are not electrically active at the lower glucose concentration indicates that the concentration-dependent stimulation of insulin secretion involves recruitment of β-cells that were previously not active. Application of GABA depolarized the cell to approximately −55 mV. This effect was blocked by the GABA_AR antagonist SR-95531. On average, puffer application of GABA depolarized β-cells from −64 ± 7 to −43 ± 4 mV (n = 4). Assuming that the membrane potential in the presence of GABA approximates E_Cl, [Cl⁻]_i can be estimated to be 32 ± 5 mmol/l (E_Cl = −60 · log[Cl⁻]_o/[Cl⁻]_i). GABA, when applied at the high concentration used in Fig. 7,A (100 μmol/l), typically triggered single action potentials. This we attribute to the large Cl⁻ conductance, which effectively clamps the membrane potential to E_Cl and thus prevents regenerative electrical activity. When a lower GABA concentration (10 μmol/l) was applied to a β-cell that was electrically active at 6 mmol/l glucose (Fig. 7 B), a smaller membrane depolarization accompanied by increased action potential firing was observed. In a series of five experiments, 10 μmol/l GABA increased β-cell action potential frequency approximately fivefold, from 0.6 ± 0.2 to 2.9 ± 0.4 Hz (P < 0.05).

Puffer application of GABA also depolarized the membrane potential of δ-cells (supplementary Fig. 7). In a series of four experiments, the δ-cell membrane potential averaged −28 ± 5 mV after application of 100 μmol/l GABA, corresponding to a [Cl⁻]_i of 54 ± 9 mmol/l. The membrane potential of α-cells was not affected by GABA, reflecting the low GABA_AR density in these cells (not shown).

Human pancreatic islets contain concentrations of GABA similar to those found in the brain (30), but the physiological function of pancreatic islet GABA in humans remains poorly defined. Here we have investigated the role of GABA and GABA_AR Cl⁻ channels in paracrine and autocrine signaling in human pancreatic islets.

Blocking GABA_ARs inhibited insulin secretion from human islets induced by 6 mmol/l glucose, i.e., close to the concentration at half-maximal stimulation (EC₅₀) for glucose-stimulated insulin secretion (31). These findings imply that GABA, by activating GABA_AR, stimulates β-cell secretion. This may seem paradoxical, since GABA is best known as an “inhibitory neurotransmitter.” GABA_ARs are ligand-gated Cl⁻ channels, and opening of these channels shifts the cell's membrane potential toward E_Cl. The effect of GABA will accordingly depend on the membrane potential of the cell and [Cl⁻]_i. Our measurements of E_Cl indicate that [Cl⁻]_i in human β-cells is ∼32 mmol/l. This value is in close agreement with that measured in mouse β-cells using a Cl⁻-sensitive fluorescent indicator (34 mmol/l [32]). Assuming a plasma Cl⁻ concentration of 110 mmol/l, E_Cl in β-cells is greater than −40 mV and thereby positive to the threshold for action potential firing in human β-cells (−55 mV [19]). For comparison, [Cl⁻]_i in neurons is <10 mmol/l, resulting in an E_Cl more negative than −60 mV (33), whereas it will produce depolarization in β-cells exposed to nonstimulatory glucose concentrations. A stimulatory role for Cl⁻ conductances in β-cells is also supported by the observation that a reduction of [Cl⁻]_i below physiological levels suppresses glucose-induced electrical activity in mouse β-cells (34). A positive GABAergic feedback loop may account for bursts of TICs observed during glucose stimulation (Fig. 4 B, ii): the release of one GABA-containing vesicle will depolarize the β-cell and this in turn (via stimulation of electrical activity) triggers the release of further vesicles. Desensitization of GABA_ARs eventually terminates the burst.

GABA_AR inhibition reduced insulin secretion at 6 but not 10 and 20 mmol/l glucose. This is similar to results obtained in INS-1 cells where GABA stimulates insulin secretion at 2.8 mmol/l glucose but is inhibitory at 28 mmol/l glucose (35). These divergent effects of GABA can be explained if increasing glucose concentrations lead to a progressive depolarization of the β-cell membrane potential (36,37). At high glucose levels, when the membrane potential approaches E_Cl, opening of Cl⁻ channels may have no further depolarizing effect and even reduce excitability (by clamping the membrane potential to E_Cl).

It has previously been demonstrated that glucose regulates GABAergic signaling in islets by increasing GABA_AR subunit expression (6) and via insulin-induced trafficking of GABA_ARs to the cell surface (8). Here we provide the important additional observation that glucose stimulates vesicular release of GABA from β-cells (Fig. 4 C). GABA was co-released with the LDCV marker serotonin, and electron microscopy demonstrated accumulation of GABA in a subpopulation of insulin granules. This is in agreement with similar observations in rat β-cells (23). The findings suggest that vesicular GABA release in human β-cells is principally due to exocytosis of insulin-containing LDCVs, although a contribution of synaptic-like microvesicle exocytosis cannot be excluded (22).

Rat β-cells release 25% of their GABA content per hour independently of the glucose concentration (38). This cannot be explained by release of GABA solely by exocytosis of LDCVs. Indeed, in cell clusters overexpressing GABA_AR, SR-95531 blocked a component of the holding current at 1 mmol/l glucose, indicating extracellular accumulation of GABA in the absence of detectable vesicular GABA release (23). Here we demonstrate that this also occurs in human β-cells. These data suggest that in addition to the glucose-dependent vesicular route of GABA release, a tonic glucose-independent background release mechanism exists in β-cells. Analysis of the excess noise (blockable by SR-95531) suggests that background release of GABA is only sufficient to activate individual channels. This indicates that background release of GABA is of a nonvesicular nature, the details of which remain to be elucidated but may involve GABA transporters operating in reverse or other nonvesicular routes (39).

Patch-clamp experiments revealed that human δ-cells express high levels of GABA_AR and that activation of these receptors—similar to what was observed in β-cells—depolarizes δ-cell membrane potential. In line with an excitatory effect of GABA_AR activation, blocking GABA_AR reduced somatostatin secretion at 3 and 20 mmol/l but not at 10 mmol/l glucose. In agreement with the study of Xu et al. (8), the GABA_AR antagonist SR95531 also modulated glucagon secretion from human islets. The fact that so few isolated α-cells exhibited GABA-activated currents may suggest that the effects on glucagon secretion are principally mediated by paracrine effects. It should be noted, however, that 14% of the α-cells in intact islets contained detectable α_1–6 subunit immunoreactivity. This is similar to the 16% obtained by electrophysiological analysis of α-cells preincubated in the presence of insulin and only slightly lower than the 22% observed in β-cells. Thus, it is possible that the expression of GABA_ARs in isolated α-cells is artifactually low because of the loss of intra-islet insulin signaling. In control experiments, we did not find any evidence that SR-95531 interferes nonspecifically with voltage-gated channels and electrical activity in islet cells. More work is needed to explain the effects of SR-95531 on glucagon and somatostatin secretion.

In summary, our data suggest that signaling via GABA and GABA_AR stimulates insulin secretion by a positive autocrine feedback loop in human β-cells. Many important pharmacological agents target GABA_AR, and it is possible that they, given the presence of GABA_AR in β-cells, also affect insulin secretion from pancreatic islets. Indeed, treatment of epileptic patients with the anticonvulsant valproic acid (which blocks the degradation of GABA) has been reported to result in increased postprandial insulin levels (40).

This work was supported by the Medical Research Council (MRC), the Wellcome Trust, the European Union (Biosim [LSHB-CT-2004-005137] and Eurodia [SHM-CT-2006-518153]), and the Department of Health (NIHR Biomedical Research Centre's funding scheme).

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

M.Br. researched data and wrote the manuscript; R.R., M.Be., A.C., and N.W. researched data; and P.R.J. and P.R. contributed to discussion and reviewed/edited the manuscript.

We thank Dr. S. Hughes, Dr. D. Gray, and Dr. S. Cross for isolation of human islets and D. Wiggins for assistance with hormone release measurements.