INS1(832/13) (12), EndoC-βH3 (13), or dispersed human islet cells were cultured and manipulated according to the protocols and procedures described in the Supplementary Methods .

One million cells were cotransfected with 650 ng of plasmid encoding either human Haemoglobin-β (HBB) wildtype (WT or PTC⁻) or with a PTC-containing mutation (NS39 or PTC⁺) fused with Renilla (RLuc), in brief, named PTC⁻ and PTC⁺, respectively. Firefly (FLuc) plasmid (14) was used as a transfection efficiency reference. Renilla and Firefly luminescence were measured by Dual-Luciferase Reporter Assay (Promega, Hampshire, England) ( Supplementary Methods ). RLuc signals were normalised to the FLuc control in both HBB NS39, in the following named HBB(PTC⁺), and HBBWT, in the following named HBB (PTC⁻), and NMD activity was calculated by dividing the RLuc/FLuc-HBB (PTC⁻) by the RLuc/FLuc-HBB (PTC⁺) ( Supplementary Figure S1A ) (14). Experiments where the control construct RLuc/FLuc-HBB(PTC⁻) was affected by cytokines were excluded, so that the resulting NMD activity only denotes the PTC-containing HBB(PTC⁺). The transfection efficiency was tested twice and resulted in an average of 80% in INS1 and EndoC-βH3 cells, as measured by FACS analysis of cells transfected with a GFP-expressing plasmid ( Supplementary Figures S1B, C ).

One million INS1(832/13) or EndoC-βH3 cells were transfected with 650 ng of plasmids encoding UPF3A, UPF3B, or UPF3BΔ42 (15), then simultaneously with NMD activity reporter plasmids (as above, 650 ng/million cells), recounted and seeded for Western blotting, glucose-stimulated insulin secretion (GSIS), viability, apoptosis (detailed below), and NMD activity assays in relevant plates and preincubated for 48 h before treatment with cytokines as explained in the Supplementary Methods .

GPIZ lentiviral shRNA particles directed against UPF2, Upf3A, or Upf3B, and a nonsilencing shRNA (NS) as a negative control were produced using the Trans-Lentiviral shRNA Packaging System in HEK293 cells (Horizon, Cambridge, England) according to the manufacturer’s protocol ( Supplementary Methods ).

Apoptosis assays were performed in duplicate by detection of caspase-3 activity using a fluorometric (µM AMC) (or/colourimetric [µM PNA/min/mL] unless stated) assay kit (Cat No. APPA015-1KT/CASP3C-1KT, Sigma, London, England) according to the manufacturer’s protocols. Cell viability was measured by the Alamarblue assay (Cat No. DAL1025, LifeTechnologies, Renfrew, England) as previously described (11).

Using TRIZOL, 33 independent biological replicates of total RNA from the NS control and/or UPF2 KD EndoC-βH3 cells exposed to cytokines, GLT, or PBS (i.e., N = 6 of each PBS-/or cytokine-exposed NS control and UPF2 KD, and N = 4/N = 5 of GLT-exposed NS control/UPF2 KD, respectively) were extracted, treated with DNase, and precipitated with isopropanol ( Supplementary Methods ). Following the manufacturer’s recommendations, 1 µg of total RNA/per isolate was used as input for the generation of sequencing libraries using NEBNext^®Ultra-TM RNA Library-Prep (Cat No. E7770, NEB, Ipswich, MA, USA) ( Supplementary Methods ). The RNA-seq raw data underwent quality control and were mapped to the human reference genome (16) and analysed using the bioinformatic pipeline described in the Supplementary Methods .

Purified total RNA (500 ng) was used for cDNA synthesis with the SuperScript™ (Cat No. 11904018, LifeTechnologies). Real-time reverse transcriptase-quantitative PCR (RT-qPCR) was performed on 12 ng cDNA with SybrGreen PCR Master Mix (LifeTechnologies) and specific primers ( Supplementary Table S2 ) and run in an ABI Real-Time PCR Machine (Applied Biosystems, ThermoFisher Scientific, Oxford, England). The raw data was analysed through −ΔCt as described in the Supplementary Methods .

Western blotting was performed using antibodies against alpha-tubulin (1:2,000) (Cat No. T5168, Sigma), UPF2 (1:1,000) (Cat No. PA5-77128, LifeTechnologies), UPF3A (1:1,000) (Cat No. PA5-41904, LifeTechnologies), UPF3B (1:1,000) (Cat No. PB9843, Boster-Bio, Pleasanton, CA, USA) and α-1-antitrypsin (1:1,000) (Cat No. TA500375, LifeTechnologies) as described in the Supplementary Methods (17).

In 12-well plates (Cat No. 150200, Nunc, Buckingham, England), 300,000 INS1(832/13) or EndoC-βH3 cells were cultured and preincubated for 2 days. GSIS was carried out using Krebs–Ringer buffer containing 2 mM or 17 mM glucose, as described (11, 18).

Insulin concentration (ng/mL or pM) was measured using a rat insulin ultrasensitive ELISA kit (Cat No. 62IN2PEG, Cisbio, Cambridge, England) or human insulin ELISA Kit (Cat No. 90095, CrystalChem, IL, USA), respectively, according to the manufacturer’s protocol.

Data are presented as means ± SEM. Statistical analysis was carried out on raw data in cases where figures gave normalised data. Group comparisons were carried out by two- or one-way ANOVA as appropriate. Significant ANOVAs were followed by a post-hoc paired Student’s t-test with Bonferroni correction using GraphPad Prism 6.0 (La Jolla, CA, USA). A paired t-test was chosen to normalise for interpassage variability in outcome parameters. Since the experimental conditions did not allow sequential sampling from the same cell culture, parallel control and interventional plate wells were considered to be paired observations and analysed accordingly statistically. If the post-hoc paired t-test did not reveal a carrying statistical difference by ANOVA, individual paired t-tests were performed and corrected for multiple comparisons. Bonferroni-corrected p-values ≤ 0.05 were considered significant and ≤ 0.10 a trend.

We previously reported that cytokines and glucolipotoxicity differentially up- or downregulate NMD component transcripts in pancreatic β cells (11). However, whether this regulation leads to increased NMD activity remains to be elucidated. Here, we used a luciferase-based NMD reporter ( Supplementary Figure S1A ) (14) to examine NMD activity in rat INS1(832/13), human insulin-producing EndoC-βH3 cells, and primary human islets. Luciferase activity analysis showed that cytokines (Cyt; 150 pg/mL IL-1β + 0.1 ng/mL IFN-γ + 0.1 ng/mL TNF-α) significantly suppressed NMD activity by nearly 50% after 18 h, but not 6 h, of exposure in INS1(832/13) cells [ Supplementary Figures S2A (i, iii)].

We then tested the effects of cytokines on EndoC-βH3 cells and dispersed human islet cells. Cytokines [2.5 ng/mL IL-1β + 10 ng/mL TNF-α + 10 ng/mL IFN-γ, chosen from dose–response experiment shown in Supplementary Figure S2B (i)] attenuated NMD activity by 30% (p = 0.009, n = 6) and 40% (p = 0.0006, n = 6) after 18 h of exposure to EndoC-βH3 cells ( Figure 1A i) and dispersed human islet cells ( Figure 1B ), respectively. Cyt increased the luciferase signal (RLuc/FLuc) from the HBB(PTC^-) ( Supplementary Figures S2B (ii), C ) but not from the HBB(PTC^-) [ Supplementary Figures 2B (v), C (ii)] in both cell models, confirming that the NMD substrate HBB(PTC⁺) was restored due to NMD activity attenuation by Cyt.

We next examined whether cytokine-mediated suppression of NMD was consistent with an accumulation of HBB(PTC⁺) transcripts. For this, we used a forward and reverse primer set to amplify the Renilla gene and the junction of exons 1 and 2 (i.e., ensuring amplification of mature transcripts only), respectively. RT-qPCR analysis demonstrated that cytokines caused significant upregulation of HBB(PTC⁺) but not HBB(PTC⁻) mRNA levels, rendering a significant reduction of the relative PTC⁻/PTC⁺ mRNA levels in INS1(832/13) (p = 0.008) [ Supplementary Figure S2A (v, vi)] and EndoC-βH3 (p = 0.001) cells [ Figure 1A ii; Supplementary Figure 2B (iv)], which verified the suppressive effect of cytokines on NMD activity.

We also examined the effect of 25 mM glucose or GLT conditions on NMD activity. Unfortunately, these conditions significantly affected the luciferase signal (RLuc/FLuc) from the HBB(PTC⁻) control, preventing further study of the effects of metabolic stressors on NMD activity.

Taken together, these results show that cytokines suppress the activity of the NMD in a range of insulin-secreting cell types.

Whereas NMD degrades unfolded protein response (UPR)-induced transcripts in compensated ER stress, NMD is suppressed in response to pronounced endoplasmic reticulum (ER) stress to allow a full-blown UPR (19, 20). Cytokines induce a robust ER stress in pancreatic β cells, largely via nuclear factor-κB (NF-κB) activation and production of nitroxidative species that inhibit the smooth endoplasmic reticulum Ca²⁺ ATPase (SERCA) 2B pump, leading to ER calcium depletion (11, 17). We have previously shown that chemical inhibition of inducible nitric oxide synthase (iNOS) alleviated ER stress and normalised cytokine-mediated regulation of NMD components in INS1 cells (11). Therefore, we asked if cytokine-mediated reduction of NMD activity was dependent on an ER stress response in β cells. We first demonstrate that thapsigargin (TG), a noncompetitive inhibitor of SERCA (21) and ER stress inducer (22), inhibited NMD activity by 50% in EndoC-βH3 cells, as measured by luciferase assay [ Supplementary Figures S3A (i, ii)]. Compared to untreated EndoC-βH3 cells (Unt), cytokines significantly augmented the increase in mRNA levels encoding the ER stress markers BiP, Xbp1, and Chop (FDR < 0.05) measured by RNA-sequencing analysis ( Figure 2A i) and later verified by RT-qPCR examination ( Figure 2A ii–vi). Finally, compared with Unt, cytokines significantly decreased the NMD activity by 30%, and this effect was counteracted by the protein kinase R-like endoplasmic reticulum kinase (PERK) phosphorylation inhibitor GSK157 (8 µM) and by the inositol-requiring enzyme 1 (IRE1α) endoribonuclease inhibitor 4µ8C (16 µM) in EndoC-βH3 cells [ Figure 2B i; Supplementary Figures S3B, C (i, ii)]. RT-qPCR analysis of the relative PTC⁻/PTC⁺ mRNA levels in EndoC-βH3 cells confirmed the NMD activity data [ Figure 2B ii, Supplementary Figure S3C (iii)].

Taken together, these results demonstrate that inhibition of UPR antagonises the cytokine-mediated reduction of NMD activity in EndoC-βH3, indicating that cytokine-mediated inhibition of NMD activity is UPR-dependent.

Since we observed in our previous study that cytokine-induced ER stress downregulated UPF3B expression in human and rodent β cells, as recovering nitroxidative-driven ER stress using the inducible nitric oxide synthase (iNOS) inhibitor N-methyl-l-arginine (NMA) (11), since transcripts encoding UPR components are NMD targets and have been shown to be stabilised by UPF3A/B depletion (19), and since UPF3B is a NMD activator in mammalian cells (23), which led to the proposal of UPF3-dependent and UPF3-independent branches of the NMD pathway (4, 10, 24). We reasoned that UPF3 regulated NMD activity in β cells. We therefore first measured the UPF3A/B expression level and then investigated the functional impact of overexpressing UPF3A/B on cytokine-mediated suppression of NMD activity in β cells. RT-qPCR examination showed that cytokines significantly downregulated UPF3B mRNA levels after 18 h in both EndoC-βH3 ( Figures 3A i, ii) and INS1(832/13) [ Supplementary Figures S4A (i, ii)] as previously reported (11). Immunoblot analysis verified overexpression of UPF3A, UPF3B, and the UPF3B dominant negative UPF3BΔ42 in both EndoC-βH3 [ Supplementary Figure S4B (i)] and INS1(832/13) [ Supplementary Figure S4C (i)]. Cytokines reduced NMD activity, and overexpression of UPF3B significantly attenuated this reduction in EndoC-βH3 [ Figure 3B ; Supplementary Figure S4B (ii)] and to a lesser extent in INS1(832/13) [ Supplementary Figures S4C (ii, iii)]. Neither UPF3A nor UPF3BΔ42 overexpression counteracted cytokine-attenuated NMD activity.

This result suggests that cytokines reduce NMD activity in β cells through downregulation of UPF3B expression.

The above findings provide evidence that the UPF3-dependent branch of NMD is involved in cytokine-mediated suppression of NMD activity. Therefore, we next investigated the impact of UPF3A/B overexpression on cytokine-induced cell death and insulin secretion. While UPF3A or UPF3B overexpression increased basal cell death, it also exacerbated the cytokine-induced apoptosis in EndoC-βH3 cells ( Figures 4A i, ii). In INS1(832/13) cells, neither UPF3A nor UPF3B overexpression changed cell viability in the absence of Cyt exposure, but UPF3B overexpression significantly aggravated cytokine-induced cell death as measured by Alamarblue and caspase-3 activity assays [ Supplementary Figures S5A (i, ii)]. Therefore, we next explored the impact of UPF3A or UPF3B deficiency on β-cell viability. Lentiviral shRNA-mediated knockdown of UPF3A and/or UPF3B [ Supplementary Figures S5B (i, ii)] significantly reduced basal INS1(832/13) cell viability ( Supplementary Figure S5C (i–iv). Taken together, these data indicate that genetic manipulations of UPF3A/B could possibly be detrimental to the β-cell viability.

In contrast to Cyt, which downregulates UPF3B in INS1 (11), EndoC-βH3 ( Figure 3A ii), and INS1(832/13) cells [ Supplementary Figure S4A (ii)], GLT does not downregulate UPF3B (11).We therefore explored the effect of UPF3 deficiency on glucolipotoxicity-induced cell death in β cells. Measurements of caspase-3 activity demonstrated that both UPF3A and UPF3B knockdown rendered a slight but significant protection against 24 h glucolipotoxicity in INS1(832/13) cells in comparison with untreated cells [ Supplementary Figure S5D (i, ii)]. Furthermore, treatment with the NMD activator Tranilast dose-dependently sensitised to glucolipotoxicity-, but not cytokine-induced, EndoC-βH3 cell death, measured by caspase-3 activity assay [ Supplementary Figure S4E (i, ii)]. Neither UPF3A nor UPF3B overexpression affected GSIS in EndoC-βH3 ( Figure 4B i, ii) or INS1(832/13) [ Supplementary Figure S5F (I, ii)] cells. Nonetheless, UPF3B overexpression profoundly lowered insulin content in EndoC-βH3 cells ( Figure 4B iii). In contrast, knockdown of UPF3A or UPF3B significantly decreased the stimulatory index, as well as provoking a substantial increase in insulin content in control INS1(832/13) cells [ Supplementary Figures S5G (ii, iii), H (ii and iii)], without altering ins1 or ins2 mRNA expression [ Supplementary Figures S5G (iv, v), H (iv, v)].

Taken together, these findings reveal that UPF3 overexpression induces basal cell death and exacerbates cytokine-mediated toxicity in β cells.

Next, we investigated the effect of UPF2 deficiency on the viability and insulin secretion of β cells because (i) UPF3A and UPF3B are involved in regulating UPF2, a key core NMD activator in mammalian cells (25, 26) by sequestering away from and bridging the exon-junction complex (EJC) with UPF1 and UPF2, respectively, leading to the NMD activation (23), and genome-wide association (GWAS) data reveal that the UPF2 variant rs145580445 is significantly associated with type 2 diabetes risk (7). We, therefore, knocked down the UPF2 gene in EndoC-βH3 cells using RNA interference and chose the three cell lines in which UPF2 was most efficiently knocked down (KD) [ Supplementary Figures S6A (i, ii)]. Examination of NMD activity using the luciferase-based NMD reporter revealed that UPF2 KD profoundly reduced NMD activity in untreated and cytokine-treated EndoC-βH3 cells [ Supplementary Figures S6B (i, ii)]. Compared with NS control, UPF2 KD slightly but significantly prevented cytokine-induced cell death ( Figures 5A i, ii). UPF2 KD had no effect on the GSIS but significantly increased insulin content ( Figures 5B i–iii).

These data indicate that UPF2 plays a crucial role in cytokine-induced β-cell apoptosis. In addition, the increase in insulin content in UPF2-deficient EndoCβH3 cells implies that insulin transcripts could possibly be targets of the UPF2-dependent NMD pathway branch.

Consistent with our observations above ( Figure 5 ), we previously reported (11) that the deficiency of SMG6, an endoribonuclease and a key effector of NMD, rendered protection against cytokine-induced cell death and was associated with increased insulin content. Therefore, we aimed to identify potential NMD target transcripts by using RNA-sequencing to assess the transcriptome of cytokine- or PBS-treated EndoC-βH3 cells stably transfected with a nonsilencing shRNA (NS) or the specific shRNA (shRNA-1 named U1) against UPF2. Due to the differential effects of Cyt and GLT on UPF3 expression cf. above, we also performed RNA-sequencing after UPF2 KD versus NS control EndoC-βH3 exposed to GLT compared with PBS-treated.

The RNA-seq datasets from either UPF2 KD or NS control EndoC-βH3 cell lines exposed to cytokines and or GLT were dimensionally reduced by principal component analysis (PCA) into two main principal components, PC1 and PC2 (p < 0.05). The PCA of the NS control EndoC-βH3 cells demonstrated a high similarity between the biological replicates, a small within-group variance, and a distinct clustering of the untreated cytokine and GLT groups [ Supplementary Figures S7A (i, ii)]. Pearson’s correlation (p < 0.05) between samples justified the clustering of biological replicates of cytokines, GLT, and untreated conditions [ Supplementary Figures S7B (i, ii)]. PCA revealed that the UPF2 knockdown increased the majority of variance in transcript isoforms of the cell, which led to patterns that visually dispersed biological replicates from the cytokine-exposed isolates, while decreasing the variances from the GLT and untreated biological replicates, which clustered them together.

A Venn diagram of the RNA-seq datasets demonstrated an approximate total of 14,000 commonly expressed transcripts (FDR < 0.05) and those differentially expressed transcripts regulated by PBS (i.e., untreated control), cytokines, or GLT [ Supplementary Figures S7C (i, ii), D (i–iv)]. Among them, cytokines regulated 54% (up) and 46% (down) ( Figure 6A i), whereas GLT impacted 48% (up) and 52% (down) of significantly expressed transcripts in NS control EndoC-βH3 cells ( Figure 6A iv). UPF2 KD changed the cytokine-mediated regulation of significantly expressed transcripts by 59% (up) and 41% (down) ( Figure 6A ii), whereas it did not change the regulation of significantly expressed transcripts by GLT (up: 47%, down: 53%) in EndoC-βH3 cells ( Figure 6A v). This indicates that UPF2 KD possibly alters the mRNA levels of identified transcript species.

To identify UPF2 KD-regulated transcripts possibly providing protection against cytokine-induced cytotoxicity, we interrogated cytokine- and/or GLT-regulated transcripts that were differentially expressed (p < 0.05) in UPF2 KD versus NS control EndoC-βH3 cells using gene enrichment analysis (GEA). Gene Ontology (GO) and Kyoto Encyclopaedia of Genes and Genomes (KEGG) gene enrichment analyses demonstrated that GLT significantly (p < 0.05) regulated transcripts in the cellular functions of RNA splicing, mitochondrial inner membrane, and purine nucleoside metabolism [ Supplementary Figure S7E (i, ii)], although these were not affected by UPF2 KD ( Figure 6B i). In contrast, GEA revealed that in both untreated and cytokine-treated EndoC-βH3 cells, UPF2 KD significantly regulated transcripts encoding proteins involved in synaptic transmission, extracellular matrix, basolateral plasma membrane, receptor complex, synaptic membrane, transcriptional repressor activity, and enzyme inhibitor activity [ Figures 6B iii, C ; Supplementary Figures S7E (iii, iv)].

RT-qPCR confirmed the logarithmic fold change of UPF2 KD-regulated transcripts in the cytokine-treated EndoC-βH3 cells versus cytokine-treated NS control cells ( Figure 6D ). The role of many of these transcripts in cell viability or insulin secretion was previously identified in pancreatic β cells. Among them, α-1-antitrypsin (1-AT) has been proposed as an antagonist against cytokine-induced pancreatic β-cell death (27, 28). Our expression profiling verified that cytokines upregulated α-1-antitrypsin, and this effect was further potentiated by UPF2 KD. To explore the potential importance of these changes, we knocked down α-1-antitrypsin using specific siRNAs in INS1(832/13) and EndoC-βH3 cells, as confirmed by quantitative WB [ Supplementary Figure S8 (Fig.7B-i, ii and iii)]. The effect of siRNA-mediated α-1-antitrypsin knockdown was inconclusive in EndoC-βH3 cells ( Figure 7A i, ii) [ Supplementary Figures S7B (i, ii)]. Both α-1-antitrypsin siRNAs aggravated cytokine-induced cell death in comparison with NS control in INS1(832/13) cells ( Figures 7A iii, iv). Compared with the NS control, α-1-antitrypsin knockdown reduced the GSIS index but had no effect on insulin content in INS1(832/13) cells ( Figures 7B i, ii, iii), possibly due to the reduced cell number.

Cytokines reportedly upregulate > 30 splicing factors, affecting alternative splicing of 35% of genes in the human islet transcriptome (6). We examined RNA-seq datasets for alternative splicing (AS) isoforms [ Supplementary Figure S9 (i)] driven by cytokines or GLT versus untreated in the NS control and UPF2 KD EndoC-βH3 cells. Among 2,123 and 2,106 cytokine-driven AS isoforms, skipped exon (SE) isoforms constituted 70.89% (p = 0.1, n = 6) and 72.5% (p = 0.1, n = 6) in NS control and UPF2 KD cells, respectively [ Supplementary Figures S9 (ii, iii)]. In contrast, 220 and 133 GLT-driven AS isoforms were identified in NS control and UPF2 KD, respectively [ Supplementary Figures S9 (iv, v)]. This differential regulation could possibly provide a reliable measure for cytokines and the role of GLT in inducing AS isoforms in β cells.

Taken together, the above transcriptome analysis of EndoC-βH3 cells indicates that cytokines increase 1–AT expression, and this is synergised by NMD attenuation.

We used a luciferase-based NMD reporter based on two separate PTC⁻ and PTC⁺ constructs whose labelled luciferase is separately measured. Thus, a yet-to-develop NMD activity reporter by which transcript RNA, protein, or their corresponding labelled luciferase activity of both PTC⁻ and PTC⁺ transcripts could be examined in one cell rather than (potentially) two separate cells will remove the limitation of the current reporter based on the transfection of the constructs into two separate cells. Moreover, the constant overexpression of UPF3A and UPF3B may result in cell death as β cells cannot cope with the overwhelming levels of these proteins above the basal level. This could explain why UPF3B overexpression reduces the basal cell viability. Further studies should (1) conduct in vivo experiments to validate the findings observed in vitro and determine if the regulation of NMD in pancreatic β cells is consistent across different cell types and conditions; (2) investigate the mechanisms underlying the regulation of NMD in pancreatic β cells, including the role of specific NMD components and the impact of different stressors on NMD activity; and (3) explore the potential therapeutic implications of targeting NMD in the treatment of inflammatory stress in β cells, including the development of novel drugs or therapies that modulate NMD activity.

The findings report NMD involvement in the development of islet autoimmunity and the destruction of pancreatic β cells in type 1 diabetes, as well as islet inflammation in type 2 diabetes. The identification of novel targets arising from cytokine-driven NMD attenuation could possibly suggest new biomarkers to monitor disease progression and may also guide the development of protein-based vaccines or antisense mRNA therapeutics in individuals who are at risk of diabetes development and or other inflammatory and autoimmune disorders.