A Phenotypic Screen Identifies Calcium Overload as a Key Mechanism of β-Cell Glucolipotoxicity

Diabetes is caused by inability of the pancreas to meet metabolic demand for insulin due to a shortage of functional insulin-secreting β-cells (1). In type 2 diabetes (T2D), elevated levels of circulating glucose and fatty acids contribute to insulin resistance in peripheral tissues, which leads to an increased demand for insulin as well as direct effects on the β-cell, such as stress and apoptosis. This concept of toxic energy excess, termed glucolipotoxicity (GLT), has been extensively studied in the last two decades (2,3).

GLT conditions are generally associated with increased oxidative and endoplasmic reticulum (ER) stress, calcium dysregulation, and inflammasome activation (4–13). In the pancreas, GLT conditions result in increased β-cell death, low insulin content, and reduced glucose-stimulated insulin secretion (GSIS) (14–17). GLT conditions have been shown to lead to reduced insulin gene expression via dysregulation of its upstream transcription factor PDX-1 (3). Decreased PDX-1 mRNA levels and nuclear exclusion of PDX-1 were observed in β-cells under GLT conditions and in islets from humans and rodents with diabetes (18–21). ER stress and ER calcium dysregulation have also been implicated in GLT-induced β-cell dysfunction (22). Free fatty acids can induce calcium influx from the ER in β-cells via activation of Gq-coupled receptors such as G-protein–coupled receptor (GPR) 120, GPR40, and inositol 1,4,5-trisphosphate receptor (IP3R). High glucose stimulation of β-cells results in membrane depolarization, leading to calcium influx through the voltage-gated L-type calcium channels on the plasma membrane and, as a result, calcium-induced calcium release from the ER via the ryanodine receptor and IP3R. The influx of calcium into the cytosol is necessary for the glucose-stimulated insulin release, but excess stimulation with high levels of glucose and fatty acids depletes ER calcium stores, which can also contribute to the reduced insulin secretion under GLT (23). Moreover, several recent reports support a role for cytosolic calcium overload in pathogenesis of diabetes. In mice expressing a mutant form of Abcc8, a key component of the β-cell K_ATP channel showing constant depolarization similar to that observed with nutrient overload, intracellular calcium is constantly elevated, leading to diabetes (24). Furthermore, mouse islet cells exposed to diabetic serum show hyperactivation of L-type calcium channels (25).

To date, there are no efficient therapies protecting patients with diabetes from β-cell death and loss of β-cell mass. Glucagon-like peptide 1 (GLP-1) signaling has been reported to exert direct β-cell–protective effects (26,27). However, despite many years of research, whether the beneficial effects of GLP-1–based therapy on β-cell mass observed in animal studies are applicable to humans is still unclear (28).

Here we report discovery of a chemical series that protects β-cells and islets from GLT-induced apoptosis and dysfunction. Mechanism of action (MOA) studies revealed that this series protects β-cells by inhibiting GLT-induced calcium influx. Our work suggests a central role for cytosolic calcium overload in β-cell GLT.

For more details including cell culture, materials and screening, please see Supplementary Methods. The GLT apoptosis assay was performed in INS1E cells as follows. After overnight incubation in low-serum culture media (RPMI plus 0.5% BSA), cells were preincubated with compounds for 1 h and then challenged with BSA-conjugated palmitate (800 μmol/L palmitate; palmitic acid [PA]:BSA ratio 6:1) (#P9767; Sigma-Aldrich). Apoptosis was measured by caspase 3/7 Glo (CaspGlo, #G8091; Promega) 24 h after the palmitate challenge. Cell viability was measured by Cell Titer Glo (#G7570; Promega) 48 h after the palmitate challenge, according to the manufacturer’s instructions, and luminescence was detected by EnVision Plate Reader (Perkin Elmer). For both CaspGlo and Cell Titer Glo assays, the percentage reduction and percentage protection, respectively, were calculated using relative luciferase units (RLU) as [(RLU of DMSO control + PA − RLU of compound + PA)/RLU of DMSO + PA]∗100%.

For TNF-related apoptosis-inducing ligand (TRAIL) apoptosis assays, Jurkat cells (#TIB-152; ATCC) were treated with compounds for 1 h, followed by treatment with human recombinant TRAIL protein (100 μmol/L final concentration) (#375-TL-010; R&D Systems), and apoptosis was measured as described above using CaspGlo kit 24 h after the TRAIL challenge.

For high-throughput screening, the final BSA-palmitate concentration was 1 mmol/L (PA:BSA ratio was maintained at 6:1).

Rat pancreatic islets were isolated from Sprague Dawley rats after perfusion with Liberase TL Research Grade (#5401020001; Sigma-Aldrich) as previously described (29). For apoptosis evaluation, islets were dispersed to single cells using a papain dissociation kit (#LK003150; Worthington) and plated in laminin-5–coated plates. After 48 h recovery, islet cells were cotreated with GLT media (0.5% fatty acid-free BSA, 500 μmol/L palmitate in RPMI, PA:BSA ratio 6:1, 16 mmol/L glucose) and compounds of interest for 72 h. For determination of cell death, islet cells were stained with 1:1,000 Hoechst, 1:1,000 propidium iodide (PI), and 1:20 annexin V Alexa Fluor 488. The staining was analyzed using the Cellomics ArrayScan (Cellomics, Pittsburgh, PA). See the Supplementary Data for additional details and calculations of apoptosis protection.

Human islets were obtained from Prodo Laboratories (Aliso Viejo, CA), which provided donors’ demographic and clinical data, HbA_1c level, and islet isolation parameters. Human islets were cultured for at least 24 h in PIM(S) human islet-specific medium (Prodo Laboratories), containing 5% human AB serum and 5.8 mmol/L glucose before assay.

For GLT-GSIS assays, rat or human islets were incubated in high-glucose, palmitate-containing media (300 μmol/L PA:BSA, 27 mmol/L glucose) overnight at 37°C in 5% CO₂. Islets were washed with Krebs-Ringer bicarbonate buffer (KRBB) containing 2.8 mmol/L glucose, and three size-matched islets per well were picked into a 96-well plate by stereomicroscope observation. For each experiment, four to eight wells (three islets each) per condition were assayed. After 2 h of incubation in KRBB at low glucose (2.8 mmol/L) condition, islets were stimulated with high glucose (16.7 mmol/L) for 1 h. The culture media were filtered to remove cell debris and stored at −20°C for analysis. Intracellular insulin was extracted overnight in lysis buffer (Cyquant kit, #C7026; Invitrogen) at −20°C. Insulin concentration in culture media and islet lysates was determined using the Cisbio insulin ultrasensitive assay kit (62IN2PEG; Cisbio) and/or the Mercodia insulin ELISA kits (rat: Mercodia #10-1,124-10; human: Mercodia #10-1,113-01) with appropriate sample dilutions. All GSIS experiments were performed three times or more. For rat islets, pancreata from eight rats were pooled and combined for each islet isolation. For human islets, islets from two donors without diabetes were used individually. For human islet assays, secreted insulin and insulin content were normalized by DNA content in human islets lysates (Cyquant DNA proliferation kit, #C7026; Invitrogen). We used GraphPad Prism 8.1.2 software for data analyses and ordinary one-way ANOVA for statistical significance assessment.

Isolated islets from eight rats were divided into seven groups (∼1,000 islet equivalents/sample) as follows: two control groups cultured in regular media (RPMI with 11 mmol/L glucose) with DMSO for 3 and 24 h, and two treated with GLT media (27 mmol/L glucose and 300 μmol/L PA:BSA in RPMI) and DMSO, or GLT media with 300 nmol/L compound D or compound E for 3 and/or 24 h. This experiment was performed three times with different islet preparations to obtain three independent biological repeats. Total RNA was prepared using the RNeasy kit (Qiagen). RNA quantity and quality were assessed using an Agilent 2100 Bioanalyzer and 1 μg total RNA was sent to the Beijing Genome Institute for reverse transcription, library preparation, and sequence analysis (Hi-Seq2000; Illumina). RNA sequencing (RNAseq) reads (100 base pairs) from paired-ends were mapped to the rat genome (rn5) with TopHat2 (version 2.0.3) and Bowtie2 (version 2.0.0) (30,31). Gene expression values were summarized as raw counts by running high-throughput sequencing (HTSeq) software (32). The variance-stabilizing transformation from DESEq software (33) was applied to the raw sequencing count data before statistical analyses. See Supplementary Data for the statistical models to identify significant gene expression changes. For PDX-1 gene expression analyses by quantitative PCR we used the TaqMan probe set Rn01423448_m1 from Thermo Fisher.

For calcium flux assay, we used a Fluorescent Imaging Plate Reader (FLIPR) Tetra High-Throughput Cellular Screening System. Papain-dissociated islets were plated on laminin-coated 384-well plates (354663; BD). After 48 h of recovery, islets were pretreated with compounds in GLT media (16 mmol/L glucose, 500 μmol/L PA:BSA in RPMI) for 1 h at 37°C in 5% CO₂. FLIPR calcium dye 6 was then resuspended in GLT media and added with appropriate compound dilutions for 2 h at 37°C in 5% CO_2. Calcium influx was measured for the next 10 min at 1-s intervals. For chronic calcium load measurements, readings were taken every minute over 10 consecutive hours after GLT treatment, and the cumulative calcium signal was calculated based on the sum of individual measurements per assay condition. Analyses were done using Molecular Devices software, with the amplitude of the response (Fig. 6A, ΔF/F maximum-minimum fluorescent signal) as the main reported readout.

Measurements of compound activities on L-type calcium channel current was performed on the IonWorks Barracuda system (Molecular Devices, LLC, San Jose, CA) by ChanTest Corporation (Cleveland, OH). Cav1.2 channel (cloned Cav1.2/β2/α2δ1) was expressed in CHO cells (ATCC), and stable cell lines were generated. For Cav1.2 measurements, CHO cells stably expressing Cav1.2 were treated with compounds diluted in a calcium-free HEPES-buffered physiological saline (composition in mmol/L): NaCl, 137; KCl, 4.0; MgCl₂, 2.8; HEPES, 10; glucose, 10; and BaCl, 5.0 (pH 7.4). The current recordings were performed on IonWorks Barracuda at ambient temperature, before compound application to the cells (baseline) and 5 min after the application. Before digitization, the current records were low-pass filtered at one-fifth of the sampling frequency (10 kHz). Cav1.2 current was measured using a stimulus voltage pattern consisting of a depolarizing test pulse to 10 mV for 300 ms from a −90 mV holding potential. Peak current was measured during the step to 0 mV. Data acquisition and analyses were performed using the IonWorks Barracuda system operation software (version 2.0.2). The decrease in current amplitude after test article application was used to calculate the percentage block relative to control. Results for each compound concentration (n ≥ 2) were averaged, and mean and SD values were calculated and used to generate dose-response curves.

The blocking effect was calculated as: %Block = (1 − I_cpd/I_Control) × 100%, where I_Control and I_cpd are the currents measured before addition of compound and in the presence of a compound, respectively. Concentration-response data for inhibitors were fit to an equation of the following form: %Block = %VC + {(%PC − % VC)/[1 + (cpd/IC₅₀)N]}, where cpd indicates compound concentration, N is the Hill coefficient, %VC is the percentage of the current run-down, and %Block is the mean value of the percentage of ion channel current inhibited at each concentration of compound. Nonlinear least squares fits were solved with the XLfit add-in for Excel 2003 software (Microsoft, Redmond, WA).

The RNAseq data reported in this study are available in the Gene Expression Omnibus repository (GSE#134272). No applicable resources were generated or analyzed during the current study.

To identify compounds that protect β-cells from GLT-induced apoptosis, we designed a low-molecular-weight compound phenotypic screen (Fig. 1A). Caspase 3/7 activation was measured in INS1E cells cultured in the presence of glucose and palmitate (GLT conditions) to identify compounds that reduced GLT-induced apoptosis. A total of 312,000 compounds were screened in 1,536-well plate format. The average Z′ for the primary high-throughput screen was 0.59, and the average RZ′ was 0.72. To verify that reduced caspase activation was not a result of cell death, compounds from the primary screen were tested in a general cell viability assay that measured ATP levels in GLT-treated INS1E cells. Compounds that reduced caspase activation but did not cause general cytotoxicity were further profiled in an apoptosis assay in Jurkat cells to eliminate general apoptosis inhibitors. Six scaffolds (chemical classes) that selectively protected β-cells from GLT-induced apoptosis remained and were further profiled in a primary rat islet cell apoptosis assay to select compounds that protect primary islet cells. Multiple readouts were used to measure apoptosis and general cell death in islets (annexin 5 and PI staining, respectively), to ensure specificity of the observed effects. Finally, compounds were tested for prevention of GLT-induced reduction of GSIS.

Compound A (Fig. 1B), a racemic mixture, inhibited GLT-induced caspase activation in INS1E cells but did not inhibit Jurkat cell apoptosis (Fig. 2A) and was selected for further characterization. The racemic compound A was separated into two enantiomers, more active B and less active C (Supplementary Fig. 1), and enantiomer B was further optimized. As a result of medicinal chemistry efforts, compound D was developed; it protected pancreatic islets from GLT-induced apoptosis at nmol/L concentrations (Fig. 2B and C). Further, a structurally related inactive tool, compound E (Fig. 2B), was used as a negative control in follow-up MOA experiments.

To determine whether the islet cells that were protected from apoptosis maintained their function, we tested whether compounds B and D could promote insulin production and secretion in response to glucose in rat and human islets compromised by GLT. As previously reported (18), a significant reduction of insulin content and insulin secretion after GLT treatment was observed in rat islets. Treatment with compounds B or D rescued both reduction in insulin content and GSIS in GLT-stressed rat islets (Fig. 3A and B). Exendin-4, a clinically used GLP-1 receptor agonist and a known stimulator of insulin secretion, served as the positive control in these experiments and increased GSIS under GLT conditions (Fig. 3B). Interestingly, unlike compounds B and D, exendin-4 did not alter islet insulin content (Fig. 3A).

Compound D also protected human islets from GLT-induced dysfunction (i.e., increased insulin content and secretion under GLT) (Fig. 3C and D). Notably, compound D did not increase insulin secretion under normal culture conditions, suggesting the effects on islet function were specific to the GLT condition (Fig. 3E). Inhibition of insulin secretion occurred with compound D treatment at μmol/L concentrations and was initially attributed to scaffold toxicity (Fig. 3E). We then initiated studies to gain more insights into the bell-shaped effects of compound D on GSIS and to identify its MOA.

To determine which GLT-regulated genes might be affected by compound D, we performed an RNAseq analysis of rat islets treated ex vivo, with or without GLT and compound D, for 3 h and 24 h, respectively. The expected suppression of β-cell function/specification induction genes, such as PDX-1 and INS1, and induction of stress-response genes was observed after 24-h GLT treatment (Supplementary Tables 1 and 2). Treatment with compound D prevented close to 80% of the gene expression changes caused by a 24-h GLT treatment (Fig. 4 and Supplementary Table 3), suggesting that compound D protects islets by regulating a central node of GLT stress response. A closely related inactive compound from the same scaffold, compound E (Fig. 2B), did not reverse gene expression changes caused by GLT (Fig. 4; see Supplementary Table 4 for all gene expression changes and statistical models used in this study).

To explore the MOA of compound D, we initially tested whether compound D may target the GLP-1 receptor, which has been reported to protect β-cells and improve their function; however, compound D failed to activate GLP-1 receptor signaling (data not shown). Moreover, compound D protected islets from GLT-induced loss of insulin content, whereas exendin-4 did not (Fig. 3A), suggesting distinct mechanisms.

GLT depletes ER calcium stores and causes ER stress (13). We hypothesized that restoration of calcium homeostasis might be the potential MOA of compound D; therefore, we tested whether calcium flux modulators could phenocopy the effects of compound D in islets. We found that L-type calcium channel blockers nifedipine and verapamil reduced GLT-triggered islet apoptosis similar to compound D, whereas the calcium channel opener FPL64176 had the opposite effect, further boosting apoptosis under GLT conditions (Fig. 5A). We also tested whether modulators of ER calcium release and reuptake can mimic the effects of compound D on apoptosis. We tested the inhibitors of ryanodine receptor-mediated calcium release, ryanodine, dantrolene, and xestospongin, which inhibit IP3R-mediated calcium release. None of these three inhibitors had any effect on apoptosis under our assay conditions (Fig. 5A). Thapsigargin, which inhibits sarco-ER calcium ATPase and calcium reuptake from the cytosol into the ER, greatly increased GLT-induced apoptosis, consistent with the published role of ER stress in GLT (Fig. 5A). The calcium channel blocker nifedipine also prevented GLT-induced reduction in insulin content, and this effect was similar to and nonadditive with that of compound D. Furthermore, the L-type calcium channel opener FPL64176 blocked the protective effects of compound D, suggesting that compound D protection requires functional L-type calcium channels (Fig. 5B). Reduction in islet insulin content under GLT stress can be partly due to the suppressed expression of PDX-1 that we and others have observed under GLT. Nifedipine treatment partially prevented the reduction in PDX-1 expression, again mimicking the compound D effect (Fig. 5C).

Compounds from this scaffold were then tested for ability to directly regulate calcium flux in islet cells. Using a fluorescent cell-permeable calcium dye, we monitored cytosolic calcium in islet cells under GLT conditions (Fig. 6A). Interestingly, the GLT-induced increase in calcium influx into the cytosol was biphasic, with a sharp initial peak within seconds of GLT media addition, followed by a second sustained plateau. To explore the nature of the two calcium peaks, we pretreated islet cells with calcium modulators with known MOA. L-type calcium channel modulators and compound D dose-dependently reduced the second calcium peak (Fig. 6B and Supplementary Fig. 2), while depletion of sarco-ER calcium with thapsigargin eliminated the first calcium peak (Supplementary Fig. 2). Although the above experiments measure acute response to GLT, extended treatment of islets with GLT media led to sustained accumulation of cytosolic calcium over the course of 10 h, and compound D prevented this accumulation, reducing calcium to baseline levels (Fig. 6C). These results suggest that acute inhibition of GLT-induced calcium influx by compound D translates into chronic suppression of GLT-induced calcium overload, and this effect may be responsible for the protective effects of the scaffold. Indeed, using diverse molecules generated as part of the medicinal chemistry activities around this scaffold, we observed a tight linear association of calcium flux inhibition with apoptosis protection within the series (Fig. 6D), which suggests molecules from this series protect β-cells from GLT by modulation of calcium influx.

To determine whether compound D is a bona fide L-type calcium channel blocker, its effect on recombinantly expressed L-type calcium channel Cav1.2 was examined in CHO cells. Nifedipine robustly blocked the Cav1.2- induced response, whereas compound D had no effect (Fig. 7A). In addition, unlike compound D, which consistently increased insulin secretion under GLT conditions, enhanced GSIS under GLT conditions was not observed with nifedipine, which resulted in further worsening of GSIS (Fig. 7B). These data imply that compound D regulates calcium influx by a mechanism that is distinct from that of nifedipine.

Here we report the design, execution, and analysis of a phenotypic screen aimed to identify molecules that protect pancreatic β-cells from nutrient overload-associated GLT. We used an in vitro INS1E GLT assay to screen a library of compounds for reduction of GLT-induced apoptosis. Our goal was to identify compounds that specifically protect from GLT-induced toxicity in β-cells. We used several counterscreens to eliminate general apoptosis inhibitors and toxic molecules and identified a few selective β-cell–protective hits.

GLT is associated with reduced insulin gene expression, reduced islet insulin content, and reduced insulin secretion (2). In rats, glucose and intralipid infusion decreases insulin biosynthesis, GSIS, and expression of β-cell genes despite increased β-cell mass, suggesting that β-cell mass per se cannot always predict the functional outcome (18). We therefore assessed the effects of our hits on primary islet function under GLT stress. Several observations indicated that the islet system recapitulates β-cell stress observed in human diabetes. First, we observed a modest increase in cell death, in line with an increased apoptotic index reported for islets from humans with diabetes (34). Second, we observed the expected reduction of INS and PDX-1 gene expression and induction of ER stress genes (Fig. 4 and Supplementary Table 4). We also observed reduction of islet insulin content (Fig. 3), which is a feature of T2D (21,34,35). Finally, we observed a GLT-induced reduction in insulin secretion that, as expected, could be reversed by a GLP-1 receptor agonist (Fig. 3), an approved diabetes drug (27).

We identified a chemical scaffold that prevented apoptosis and improved insulin secretion in rodent and human primary islets under GLT stress. MOA studies pointed to the suppression of cytosolic calcium influx as a primary mechanism responsible for islet protection from GLT-induced apoptosis. We generated a series of compounds structurally related to our initial hit and observed a tight association of calcium flux modulation with apoptosis protection within the series. Unfortunately, the undesired pharmacokinetic properties prevented us from comprehensive testing of these molecules in vivo in a T2D animal model.

The discovery that our compounds inhibit calcium flux was unexpected. Although calcium dysregulation by excess fatty acids is well documented (13) and apoptosis protection has been described for calcium channel blockers (36), the increased insulin secretion we observed was contrary to the well-established inhibition of insulin secretion with calcium channel blockers. Indeed, we have not observed inhibition of recombinant L-type channel with compound D or improvement of insulin secretion with the canonical calcium antagonist nifedipine. We hypothesize that compound D regulates calcium influx by a mechanism that is distinct from that of nifedipine and is a state-dependent L-type calcium channel antagonist or represents a novel class of calcium antagonists.

The calcium antagonist described here could largely prevent gene expression changes induced by GLT, suggesting that intracellular calcium overload is the central mechanism underlying GLT-induced apoptosis and dysfunction. Several recent reports showed a link between calcium dysregulation and diabetes (37). Interestingly, we did not observe apoptosis protection for ER calcium release modulators but did observe protection with diverse L-type calcium channel blockers. The antidiabetic activity of verapamil was recently confirmed in human studies, where patients on verapamil showed decreased incidence of newly diagnosed T2D (38) and improvement in diabetes in a small clinical study (39). These human data support a role for calcium overload in diabetes pathology.

Our results are consistent with a model (Fig. 8) where GLT conditions cause an increase in cytosolic calcium influx, which can induce apoptosis as well as reduction in gene expression of critical β-cell function genes, such as PDX-1 and insulin, followed by reduction in islet insulin content and GSIS. Molecules identified in the β-cell protection screen described here can reduce the excessive calcium influx by restoring nearly optimal calcium levels and protect islets from all of the above detriments. Unlike the canonical L-type calcium channel blocker nifedipine, these molecules allow insulin secretion under conditions of calcium channel overstimulation. Calcium antagonists with such properties may prove to be a novel class of diabetes treatment and prevention drugs.

Acknowledgments. The authors thank Faye Zhao for uploading RNAseq data into Gene Expression Omnibus, Yunshan Peng, David Parker, Katsumasa Nakajima, Timothy Rasmusson, Toshio Kawanami, T.R. Vedananda, and Yongjin Gong for synthesizing analogs of compound D, Jeffrey A. Brown for insulin measurements, and Martin G. Waters and Jovita Marcinkeviciene (all from Novartis Institutes for BioMedical Research at the time of this work) for discussions and critical reading of the manuscript.

Duality of Interest. All authors were employees of the Novartis Institutes for BioMedical Research at the time of the reported studies. No other potential conflicts of interest relevant to this article were reported.

Author Contributions. J.V., J.Y., L.S., S.X.W., R.Z., S.F., C.-H.C., A.C.W., and B.D. performed experiments and analyzed data, A.C., F.Z., G.T., T.M.S, B.D., D.M.R., X.C., and A.B. contributed to discussions and data analyses. H.X. analyzed RNAseq data. A.B. wrote the manuscript. All authors reviewed and edited the manuscript. A.B. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.