Electrophysiological properties of human beta-cell lines EndoC-βH1 and -βH2 conform with human beta-cells

Limited access to human islets has prompted the development of human beta cell models. The human beta cell lines EndoC-βH1 and EndoC-βH2 are increasingly used by the research community. However, little is known of their electrophysiological and secretory properties. Here, we monitored parameters that constitute the glucose-triggering pathway of insulin release. Both cell lines respond to glucose (6 and 20 mM) with 2- to 3-fold stimulation of insulin secretion which correlated with an elevation of [Ca2+]i, membrane depolarisation and increased action potential firing. Similar to human primary beta cells, KATP channel activity is low at 1 mM glucose and is further reduced upon increasing glucose concentration; an effect that was mimicked by the KATP channel blocker tolbutamide. The upstroke of the action potentials reflects the activation of Ca2+ channels with some small contribution of TTX-sensitive Na+ channels. The repolarisation involves activation of voltage-gated Kv2.2 channels and large-conductance Ca2+-activated K+ channels. Exocytosis presented a similar kinetics to human primary beta cells. The ultrastructure of these cells shows insulin vesicles composed of an electron-dense core surrounded by a thin clear halo. We conclude that the EndoC-βH1 and -βH2 cells share many features of primary human β-cells and thus represent a useful experimental model.

Solutions. All electrophysiological experiments were performed at 32 °C in a standard or perforated patch whole cell configuration. Recordings were made using an EPC-10 amplifier and Pulse software. The action potential-like voltage-clamp commands were constructed by averaging 12 and 15 action pontential in EndoC-βH1 and EndoC-βH2, respectively and used to elicit currents. Action potentials were also recorded from human primary beta cells to allow comparison with action potential shape in the two cell lines.
For the perforated patch recordings and measurements of K + current, the extracellular medium was composed of (mmol/L) 138 NaCl, 3.6 KCl, 0.5 MgSO 4 , 0.5 NaH 2 PO 4 , 5 NaHC0 3 , 1.5 CaCl 2 and 5 HEPES (pH 7.4 with NaOH). The extracellular medium was supplemented with 1, 6 or 20 mmol/L glucose and tolbutamide (0.2 mmol/L) and diazoxide (0.5 mmol/L) as indicated. For the measurements of Na + and Ca 2+ currents and capacitance measurement of exocytosis, outward K + currents were inhibited by inclusion of 20 mmol/L tetraethylammonium (TEA) in the extracellular medium (NaCl correspondingly reduced to maintain iso-osmolarity). Na + currents were recorded after substitution of CaCl 2 for CoCl 2 (to block the voltage-gated Ca 2+ currents).
Current analysis. The net contribution of a current component to the total current was estimated by subtracting the current responses after addition of specific blockers from those observed prior to the addition. In experiments involving multiple blockers, these were applied sequentially in the continued presence of the other inhibitor(s). To compensate for variations in cell size, responses have been normalised to cell capacitance, which is proportional to cell surface area. Compared to primary human beta-cells, both cell lines appear more fragile and it was difficult to maintain the recordings long enough to allow sequential application of multiple channel blockers.
Sodium channels activation/inactivation. The voltage dependence of activation by fitting to the Boltzmann equation (Eq. 1) the increased conductance normalised to its maximum evoked from a voltage of −70 to 20 mV: where V 0.5 is the membrane potential at which activation is half-maximal and dx is the slope. Monophasic inactivation of the Na + current was fitted to a single Boltzmann equation (Eq. 2): 1 exp (2) x V dx h where V h is the voltage at which current inactivation is half-maximal, and x is the slope factor.
[Ca 2+ ] imaging. The cells were plated at the same density as that in the culture flask but in a 60 μl strip on the coverslip. EndoC-βH1 and -βH2 cells were transfected with GCaMP5G, expressed from pCMV-GCaMP5G (Addgene plasmid # 31788) using lipofectamine (ThermoFischer Scientific, Loughborough) as detailed in the Supplementary material. Prior to experiments, the coverslip was transferred to an in-house recording chamber, superfused at a rate of 60 μl/min with KRB solution at 34 °C and imaged using a 10-14x magnification on a Zeiss AxioZoom.V16 microscope (Zeiss, Germany). Synchronisation of the recording with the perfusion is achieved by addition of food colorant to the superfusion medium at the end of the experiment. Image sequences were analysed as detailed in the Supplementary material.

Immunocytochemistry.
EndoC-βH1 and -βH2 were fixed in 4% paraformaldehyde, permeabilised in 0.1% (v/v) Triton X-100, and blocked in 5% (w/v) goat serum. Immunofluorescence was performed using a guineapig anti-insulin antibody directed against human B chain (in-house, 1:1000) and anti-guineapig Texas red secondary antibody. Glucagon and somatostatin were labeled using anti-mouse glucagon antibody (Sigma, 1:1000) and anti-mouse somatostatin antibody (Santa Cruz Biotechnology, Germany, 1:200 where D is the measured averaged granule diameter, h is the smallest granule diameter and T the section thickness. From the average vesicle area, the average diameter (D) was estimated as 0.

Glucose triggers increases in [Ca
EndoC-βH1 and -βH2 cells are electrically excitable. We correlated insulin secretion and changes in [Ca 2+ ] i with electrical activity. At 1 mmol/L glucose, the membrane potential was −70 to −60 mV and many EndoC-βH1 and -βH2 cells exhibited spontaneous action potential firing. Increasing glucose concentrations to 20 mmol/L resulted in membrane depolarisation to −55 mV and stimulation of electrical activity (Fig. 2a). On average, glucose (20 mmol/L) increased action potential frequency by 30-(n = 3 cells, p = 0.03) and 25-fold (n = 4 cells, p = 0.01) in EndoC-βH1 and -βH2 cells, respectively (Fig. 2b). The characteristics of glucose-induced action potentials in EndoC-βH2 were largely similar to those in primary human beta-cells potential and peaked at 5 mV ( Fig. 2c). However, in EndoC-βH1 cells, the action potentials were generally broader and in 90% of the measured cells, they peaked at voltages above +5 mV.
Na + channels. In both cell lines, voltage-gated Na + currents (I Na ) elicited by membrane depolarisations from −70 to 0 mV were insensitive to the Ca 2+ channel blocker Co 2+ but highly blocked by tetrodotoxin (TTX, a blocker of most voltage-gated sodium channels) (Fig. 3a). Figure 3b shows families of Na + currents during depolarisations between −60 and +20 mV evoked from a holding potential of −150 mV or −70 mV. Both activation and inactivation became more rapid with stronger depolarisations. At 0 mV, activation and inactivation were complete within <1 ms and 5 ms, respectively. Figure 3c shows the current (I)-voltage (V) relationships recorded when the cells were held at either −70 mV or −150 mV. In both cell types, I Na became detectable during depolarisations to −40 mV, was maximal at 0 mV and decreased at more positive voltages with an extrapolated reversal potential at +70 mV. The maximum current elicited from −70 mV (−9.5 ± 1 pA/pF, n = 24 EndoC-βH1 cells and −19 ± 7 pA/pF, n = 11 EndoC-βH2 cells) increased by 90% and 60% respectively when the cells were held at −150 mV (Fig. 3c). This suggests that the Na + channels in EndoC-βH1 and -βH2 cells undergo partial voltage-dependent steady-state inactivation at unphysiologically negative membrane potentials, in agreement with what is observed in other insulin-secreting cells 27 .
Voltage-gated Na + channels characteristically undergo voltage-dependent inactivation. We characterised this by a two-pulse protocol (a standard test pulse was preceded by a conditioning prepulse). In EndoC-βH1 and -βH2 cells 58% and 36% of the measured I Na exhibited monophasic inactivation; half-maximal inactivation (V h ; see Eq. 2 in Supplementary material) was at −63 ± 2 mV (n = 15) and −59 ± 2 mV (n = 12) in EndoC-βH1 and -βH2 cells, respectively (Fig. 3e). The remaining cells showed biphasic voltage dependence of inactivation and in addition to the main component contained a current component that accounted for 30-40% of the total current that inactivated with V h -values of −94 ± 4 mV (n = 11) to −91 ± 3 mV (n = 21) (not shown). The latter component likely accounts for the increase in I Na amplitude when the cells were held at −150 mV rather than −70 mV, but its functional significance remains obscure, as it will be completely inactivated at physiological membrane potentials. Overall, the I Na activation and inactivation properties in EndoC-βH1 and -βH2 cells are similar to those observed in primary human beta-cells 23 Figure 3c shows I Na during an action potential-like voltage-clamp stimulation (based on the action potentials recorded in EndoC-βH2 cells). Peak I Na elicited by this stimulation paradigm were small: 3.1 ± 0.3 pA.pF −1 and 3.9 ± 1 pA.pF −1 in EndoC-βH1 and -βH2 cells, respectively (Fig. 3f).
At the mRNA level and compared to the gene set of Na + channels transcripts EndoC-βH1 and -βH2 cells express particularly high levels of SNC8A (Na v 1.6) and SNC9A (Na v 1.7). SCN3A (Nav1.3) is also highly expressed in EndoC-βH2 cells. Furthermore, SCN5A and SCN7A are expressed at relatively high levels but these encode TTX-resistant channels. Given that I Na in both cell types is TTX-sensitive, it appears that these transcripts do not give rise to functional channels. Of the β-subunits, only SCN3B is expressed at significant levels ( Supplementary  Fig. 2, and Supplementary dataset).
Ca 2+ channels. Voltage-gated Ca 2+ currents (I Ca ) were recorded in the presence of TTX. The relative contribution of P/Q-, L-and R-type Ca 2+ channels was estimated by sequential addition of ω-agatoxin, isradipine and SNX482 in presence of TTX to inhibit I Na+ (Fig. 4a,c). Figure 4b,d compares the voltage dependence of the total I Ca and the different (pharmacologically isolated) I Ca components (EndoC-βH1, n = 5; EndoC-βH2, n = 3 cells). In both cell lines, there was a 'shoulder' at voltages between −50 and −20 mV. In both cell types, P/Q-type Ca 2+ channels accounted for ~60% of the total current, whereas L-and R-type Ca 2+ channels contributed 20-30% and 5-10%, respectively (Fig. 4b,d). Similar to previous measurements on primary human beta-cells 23 , the total I Ca density was ~10 pA/pF in both cell types (Fig. 4e). Figure 4g compares the activation of the total I Ca and the respective components during an action potential in EndoC-βH2 cells. It is clear that both the L-and P/Q-type Ca 2+ channels activate during the brief action potential. Consistent with electrophysiology, transcriptomic profiling revealed high expression of CACNA1A (P/Q-type) and CACNA1C/D (L-type) Ca 2+ channels and low levels of CACNA1E (R-type) expression. Moreover, CACNA1H (T-type) Ca 2+ channels are expressed at high levels and are likely to account for the shoulder between −50 and −20 mV in the current-voltage relationships (Fig. 4a,c). Of the auxiliary subunits, high expression of CACNG4 (γ 4 ), CACNA2D (α 2 δ 2 ) and CACNB1-3 (β 1 -β 3 ) was observed (Supplementary Fig. 3; and Supplementary dataset).
K + channels. Voltage-activated K + -currents (I K ) elicited during membrane depolarisations from −70 mV to membrane potentials between −20 and +70 mV in EndoC-βH1 and -βH2 cells are shown in Fig. 5a. The amplitude of I K increased linearly with the applied voltage and the density was ~50% larger in EndoC-βH2 (n = 3) than in -βH1 (n = 4 cells) (Fig. 5b). We estimated the contribution of voltage-(K v 2) and Ca 2+ -activated (BK) channels to I K by sequential addition of their cognate blockers stromatoxin and iberiotoxin respectively (Fig. 5c). Notably, BK channels played a more prominent role in EndoC-βH2 than in EndoC-βH1 during depolarisations to −10 and 0 mV (Fig. 5d). Figure 5e shows the contribution of K V 2 and BK channels to the outward current during voltage-clamp depolarisations based on the action potentials in EndoC-βH1 and -βH2 cells. This analysis reveals that BK current component activated rapidly during the action potential and at a time when no outward current was seen in the EndoC-βH1 cells (Fig. 5e, vertical dashed line).
The transcript profile of the K channels expressed in EndoC-βH1 and -βH2 cells reveals high levels of the K V 2 channel genes KCNH6, H2, KCNB2, KCNQ2 and the BK channel genes KCNMA1 and KCNMB3 (Supplementary Fig. 4

and Supplementary dataset).
Exocytosis. Depolarisation-evoked exocytosis was monitored as increases in cell capacitance. Figure 6a,b compares the voltage dependence of exocytosis. Exocytosis was trigged by 500 ms depolarisations to voltages between −40 and +40 mV. In both cell types, exocytosis was elicited at voltages above −30 mV, was maximal at ~0 mV and declined at more depolarised voltages (EndoC-βH1, n = 6; EndoC-βH2, n = 10 cells). This voltage dependence echoes that of I Ca . The kinetics of exocytosis during depolarisation to 0 mV (i.e. close to the peak of the action potential) was determined by application of progressively longer depolarisations (Fig. 6c,d). In both EndoC-βH1 and -βH2 cells, exocytosis was elicited by depolarisations as short as 20 ms, showed an initial plateau between 50 and 100 ms and a secondary acceleration during longer depolarisations (EndoC-βH1 n = 12; EndoC-βH2 n = 14 cells). The exocytotic response to a 800 ms depolarisation once the capacitance had stabilized was 119 ± 31 fF and 181 ± 40 fF in EndoC-βH1 and -βH2 respectively (Fig. 6d, insert).
We also monitored exocytosis during repetitive stimulation consisting of ten 500 ms depolarisations to 0 mV (Fig. 6e). Figure 6f compares the responses to the individual depolarisations in EndoC-βH1 (n = 12 cells) and -βH2 (n = 14 cells). The kinetics of exocytosis was biphasic, echoing the kinetics observed in primary human beta-cells 29 . The largest response was elicited by the initial depolarisations with the first two pulses accounting for ≥50% of the total increase in capacitance and the responses during the last four pulses were small. The total increase in cell capacitance evoked by the train averaged 450 fF in both cell types (Fig. 6f, inset).
Transcriptomic analysis of genes encoding proteins involved in exocytosis revealed higher expression of SNAP25 in EndoC-βH1, than in EndoC-βH2 cells. The other SNAREs, VAMP2 and Syntaxin1A (STX1A) were expressed to the same extent ( Supplementary Fig. 5 and Supplementary dataset). Other components of the exocytotic machinery, including munc18a-c (STXBP1-3), munc13b (UNC13B) and the complexins (CPLX1-2) were also highly expressed. Of the Ca 2+ sensors of exocytosis, the high-affinity Ca 2+ sensor SYT7 was the predominant synaptotagmin in both cell types but relatively high levels of SYT1, 5 and 9 were also detected in EndoC-βH1 cells.

Subset of EndoC-βH1 and -βH2 cells are polyhormonal cells. Ultrastructural analysis revealed that
the secretory granules in EndoC-βH1 and -βH2 were of variable size and that most granules exhibited a clear thin halo and a central insulin-containing core (Fig. 7a). The granule density (granules per cytoplasm area; N A ) was slightly higher in EndoC-βH2 cells than in the -βH1 cells (Fig. 7b, n = 10 cells in both cell types, p = 0.045) but the intragranular insulin density (estimated from immunogold labelling) was correspondingly reduced (p = 9.1 × 10 −8 ) (Fig. 7c). On average, 4.5 ± 1.7% (EndoC-βH1) and 7.6 ± 2.4% (EndoC-βH2) of the vesicles per EM section were directly docked to the plasma membrane. In both cell types, the cross-sectional granule area was 0.03 µm2 (Fig. 7d; n = 265 and 337 vesicles in EndoC-βH1 and -βH2 respectively), corresponding to a granule diameter of 200 nm from which we estimate a granule surface area of 0.12 µm 2 (assuming spherical geometry).
Immunocytochemistry suggested that most EndoC-βH1 and -βH2 cells contained only insulin-labelled granules but that a small subset of cells also contained somatostatin-and/or glucagon-positive granules ( Supplementary Fig. 6). Whereas insulin-and glucagon-positive cells were found only in EndoC-βH2, somatostatin-positive cells were found in both cell lines. Electron microscopy in conjunction with immunogold labelling indicated that somatostatin and insulin could be contained in the same granules (Fig. 7e). Transcript analysis confirmed high expression of insulin (INS) in both EndoC-βH1 and -βH2 cells and that glucagon (GCG) and somatostatin (SST) were expressed at particularly high levels in EndoC-βH2 cells. There was also high expression of the somatostatin receptors 1, 2 and 3 (SSTR1-3) (Supplementary Fig. 7 and Supplementary dataset). Notably, addition of the SSTR2 antagonist CYN154806 doubled glucose-induced insulin secretion in EndoC-βH2 cells (p = 0.035) (Fig. 7f).
Transcriptome and foetal origin. The relative low stimulation of insulin secretion and the polyhormonality observed in EndoC-βH2 cells may be remaining features from their foetal origin 30  we compared the publically available transcriptomic data from foetal and adult human pancreatic beta cells with the RNA sequencing datasets we generated from EndoC-βH1 and -βH2 cell lines. We analysed a subset of genes that were consistently lowly or not expressed in human adult beta cells but highly expressed in foetal beta cells 32,33 . Such genes encoded for either components of the early development (NGN3 (Neurogenin 3)) or the inflammatory response (PTGS2 (Prostaglandin-Endoperoxide Synthase 2), S100A9 (S100 Calcium Binding Protein), and DEFA4 (Defensin Alpha 4)) 32 . Comparison of a subset of genes known to be mainly expressed in human foetal beta cells revealed that the human beta cell lines' transcriptomes are closer to the one of an adult primary beta cell ( Supplementary Fig. 9).

Discussion
We have characterised the electrophysiological properties of two human beta-cell lines: EndoC-βH1 and -βH2.
We confirmed that both cell lines are glucose-responsive with the stimulatory effect on insulin secretion limited to a 2-to 3-fold enhancement [8][9][10] . At the transcriptional level, these cell lines exhibit a greater similarity to adult than to foetal human beta cells. We acknowledge that these human beta-cell lines exhibit some variability between different batches in terms of glucose responsiveness of insulin secretion and even presence of ion channels. However, this is not unique to these cell lines and, provided appropriate controls are performed in parallel, this variability should be possible to manage. In this context it is worth remembering that also primary human islets show significant inter-preparation variability. Here we discuss a few aspects of this work that we find particularly interesting and important. We correlate our findings in the EndoC-βH1 and EndoC-βH2 cell lines to what has previously been found for primary human beta-cells.
K ATP channel activity and action potential firing. The action potentials induced by glucose originated from a membrane potential of ~−50 mV and usually peaked at +5 mV (or above). Both EndoC-βH1 and -βH2 are equipped with K ATP channels. The measured K ATP channel density at 1 mmol/L glucose in metabolically intact beta-cells is 0.05 nS/pF. As the cells had cell capacitances of 10-12 pF, these values correspond to a resting conductance of ~0.5 nS and an input resistance (the reciprocal of the resting conductance) as high as 2 GΩ. The former value is only ~10% of the corresponding value in mouse beta-cells but comparable to that observed in primary human beta-cells 1 . The high input resistance suggests that small currents (such as those resulting from the opening of individual ion channels) may produce a membrane depolarisation large enough to trigger an action potential. This probably explains why EndoC-βH1 and -βH2 cells generate spontaneous action potentials and exhibit [Ca 2+ ] i transients at 1 mmol/L glucose. Both increasing glucose (to 6 and 20 mmol/L) and application of the K ATP channel inhibitor tolbutamide reduced the resting conductance by 50-60%. Thus, it appears that the net K ATP density is ~0.03 nS/pF with the rest representing the contribution of other ion channels (or 'leak' around the recording electrode). The inhibitory effect of glucose could be reversed by diazoxide and the input conductance measured in the presence of this K ATP channel activator was 3-to 6-fold that measured in the presence of glucose alone in EndoC-βH1 and -βH2 cells, respectively.
The finding that high glucose and tolbutamide inhibit K ATP channel activity and stimulate insulin secretion similarly should not be taken to indicate that glucose solely operates by the triggering (K ATP channel-dependent) mechanism with little involvement of the amplifying (K ATP channel-independent) mechanism. On the contrary, our recordings of the cytosolic [Ca 2+ ] i show that tolbutamide stimulates a greater number of cells than glucose does, suggesting that tolbutamide exerts a weaker stimulatory effect when normalised to the number of cells activated.
Voltage-gated ion channels. We found that the EndoC-βH1 and -βH2 cells express voltage-gated ion channel counterparts very similar to that found in primary human beta-cells 23 . Thus, action potential firing in these human beta-cell lines involves activation of voltage-gated Na + , Ca 2+ and K + channels 34 . Interestingly, most of EndoC-βH1 cells exhibit large-amplitude and long-duration action potentials. Using voltage-clamp command pulses based on recorded glucose-induced action potentials, we found that this correlated with delayed and reduced activation of an iberiotoxin-sensitive K + current (reflecting the opening of large-conductance Ca 2+ -activated K + channels; BK) and slower action potential repolarisation. The same analysis also suggests that Kv2 channels play a relatively more prominent role in action potential repolarisation in these cell lines than in human primary beta cells 23 . Voltage-clamp analyses of the current responses further suggest that potential depolarisation principally reflects opening of voltage-gated P/Q-and L-type Ca 2+ channels with little contribution by voltage-gated Na + channels. This is because action potentials originate from fairly depolarised membrane potentials (−55 to −50 mV). At these membrane potentials, the Na + channels are largely inactivated, which explains why I Na contributes only marginally to action potential firing at steady-state.
Correlating electrophysiology with gene expression. To establish how authentic EndoC-βH1 and -βH2 cells are as models of primary human beta-cells, we correlated the electrophysiological data with gene expression. We found that there was, in general, good agreement at the transcript level between both cell lines and primary human beta-cells. Among the voltage-gated Na + channels, SCN8A (Nav1.6) and SCN9A (Nav1.7) are expressed at particularly high levels. We detected high expression of L-type Ca 2+ channel α-subunits CACNA1C (α1C), CACNA1D (α1D) and P/Q-type Ca 2+ channels CACNA1A. Furthermore, very high expression of the T-type Ca 2+ channel CACNA1H was detected. Expression of CACNA1E and CACNA1B (encoding R-and N-type Ca 2+ -channels, respectively) were low in EndoC-βH1 and -βH2 cells (echoing what is seen in primary human beta-cells 23,33  BK α-subunit gene KCNMA1. However, despite the evidence for reduced BK activity in EndoC-βH1 cells compared to EndoC-βH2 cells, mRNA levels were comparable between the two cell lines. Overall, the expression of these voltage-gated ion channels in EndoC-βH1 and -βH2 cells mirror that seen in primary human beta-cells 32,33 . Correlating ultrastructure, exocytosis and insulin secretion. Using immunofluorescence as well as electron microscopy, we show that both cell lines are well granulated, although the granules are not evenly distributed within the cell. The secretory granules are found at higher density in cytoplasmic projections (Supplemental . Granules are morphologically heterogeneous with a majority characterised by a very thin halo surrounding the electron dense core (insulin). The granule density (N A ) can be converted to the volume density (N V ) (Eq. 3) and corresponded to ~13 and 20 granules.μm −3 in EndoC-βh1 and βh2, respectively. When multiplied by the extranuclear cell volume, these values convert to ~6,000 and 15,000 secretory granules/cell in EndoC-βH1 and -βH2 cells, respectively. This is not too different from the numbers found in primary beta-cells 35,36 . Given that between 4 to 7% of the vesicles appeared to be directly in contact with the plasma membrane, we estimate the docked granule pool is of 250-650 vesicles, in fair agreement the 600 docked granules estimated in human primary beta cells 2,37 .
Exocytosis. The measurements of exocytosis also indicate that the late steps of insulin release largely recapitulate what is observed in primary beta-cells. However, EndoC-βH1 cells could be distinguished from the EndoC-βH2 cells by the kinetics of secretion. Whereas most of exocytosis in the EndoC-βH2 cells occurred during the actual depolarisation ('phasic exocytosis'), exocytosis in many EndoC-βH1 cells occurred after the depolarisation and when the membrane potential had returned to −70 mV ('asynchronous exocytosis'). This suggests that the kinetics of exocytosis in EndoC-βH1 mirrors the depolarisation-evoked [Ca 2+ ] i transients, which typically takes a few seconds to return to basal 38 . This is consistent with the high expression of the high-affinity Ca 2+ -sensor synaptotagmin 7 (SYT7) in EndoC-βH1 cells. However, this cannot be the sole explanation as equally high expression of SYT7 is seen in EndoC-βH2 cells and yet exocytosis was 'phasic' . Quantitatively, the exocytotic responses observed in EndoC-βH1 and -βH2 cells are impressive. A single 800 ms depolarisation produced a capacitance increase of 100 and 180 fF in EndoC-βH1 and -βH2 cells, respectively. The granule area estimated from electron microscopy was 0.12 µm 2 . With a specific capacitance of 10 fF.μm −2 , this granule area predicts that each granule should add 1.2 fF of capacitance upon fusion. Thus, the exocytotic responses equate to the discharge of 80-150 granules, which corresponds to 15-25% of the docked vesicles pool and ~1% of the total granule number. This echoes the size of the Ready Releasable Pool (RRP, ~200 vesicles) described in human pancreatic beta cells 37 . From the relationship between pulse duration and exocytosis, we estimate that a 10 ms depolarisation (the approximate duration of an action potential) will produce an exocytotic response of ≤1% of that produce by the 800 ms pulse (Fig. 5). Thus, we can estimate that each action potential will (on average) result in the release of 0.8-1.5 granules. Multiplying these values with the observed action potential frequency (1.45 Hz) suggests that EndoC-βH1 and -βH2 cells release an astonishing 100% and 50% of their total insulin content per hour. However, the [Ca 2+ ] i measurements suggest that only 20-25% of the cells respond to glucose. This implies that EndoC-βH1 and -βH2 respectively release about 20% and 10% of their insulin content per hour, in fair agreement with the secretion rates actually observed. Impact of 'polyhormonality'. At the mRNA level, glucagon and somatostatin levels equal that of insulin in EndoC-βH2. However, at the protein level only a relatively small fraction of cells are polyhormonal. Immunogold labelling in conjunction with electron microscopy revealed that these cells could store both insulin and somatostatin within the same granules. Interestingly, there was also relatively high expression of the somatostatin receptors 1, 2 and 3 (SSTR1-3), raising the possibility of autocrine inhibition. Indeed, when the cells were treated with 1 μmol/L CYN154806, glucose-induced insulin secretion was doubled. As this antagonist is selective for SSTR2, it is possible that even greater enhancement would have been observed if SSTR1 and SSTR3 had also been targeted.

Conclusion
We conclude that both EndoC-βH1 and -βH2 cells are reliable models of primary human beta-cells. We acknowledge that the secretory response to glucose and tolbutamide appears relatively limited and subjected to batch-to-batch variations. This is illustrated by the variable GSIS fold stimulation at high glucose reported in different studies 8,21,39 which could even approach that seen in primary beta-cells 16 . Electrophysiologically, both EndoC-βH1 and -βH2 cells closely mirror primary human beta-cells both at the functional and the gene expression levels. Our detailed characterisation of both cell lines demonstrates that they represent versatile models of primary human beta-cells which can be used for high-throughput pipelines 18,21 as well as detailed in vitro studies of the insulin secretory defects 19,40 . The fact that they are relatively easy to modify genetically makes them an attractive and powerful tool for future investigations of the relationship between gene expression and beta-cell function.