Cohort 1 consisted of 725 individuals (47 ± 11 years, 54% men) recruited in the northwest of Spain, including general population, and obesity and diabetes outpatient clinics in which the percentage of obese individuals was 72% and the percentage of type 2 diabetic individuals was 11% (13). Cohort 2 included 616 Caucasian subjects selected for a study of non-classic cardiovascular risk factors performed in the northwest of Spain (Asturias) (14). Participants (52 ± 12 years, 45% men, 26% obesity, 11% T2D) were randomly identified from a census and invited to participate.

Clinical characterization of human cohorts included a standardized questionnaire, physical examination and the performance of routine laboratory tests. Height and weight were measured by trained personnel using calibrated scales and a wall-mounted stadiometer, respectively, and with the participant in light clothing and without shoes. Body mass index (BMI) was calculated by dividing weight in kilograms by the square of the height in meters (kg/m²). Obesity was set at BMI≥30 kg/m². The waist of the subjects was measured with a soft tape midway between the lowest rib and the iliac crest, hip circumference was measured at the widest part of the gluteal region, and waist-to-hip ratio was then calculated. Together with clinically relevant information and subsidiary data, the number of cigarettes/day (if any) and the use of hormonal contraceptives were recorded. In those participants that agree (>75%), oral glucose tolerance test (OGTT) was performed to measure glucose tolerance. Blood samples from all the participants were collected, and after 15 minutes, tubes were centrifuged at 4,000 r.p.m. at room temperature. The serum and peripheral blood leukocytes were separated and immediately frozen at –80°C. Genomic DNA was extracted from blood samples following standard purification methods (QIAamp DNA Blood Mini Kit, Qiagen, Hilden, Germany) and DNA quantity and purity was determined using a spectrophotometer (GeneQuant, GE Health Care, Piscataway, USA). The targeted single nucleotide polymorphism (SNP) rs2076380 was genotyped by means of a predesigned rhAmp™ allelic discrimination assay (Hs.GT.rs2076380.A.1; Thermo Fisher Scientific, Massachusetts, USA) and the rhAmp Genotyping Master Mix (IDT, Coralville, USA), using a LightCycler 480 RT-qPCR System sequence detector (Roche Diagnostics, Barcelona, Spain). Replicates and positive and negative controls were included in all reactions.

The EndoC-βH1 human pancreatic cell line (Univercell Biosolutions, Paris, France) was cultured in plates coated with Matrigel-fibronectin (100 mg/ml and 2 mg/ml, respectively; Sigma-Aldrich, Burlington, USA) in Opti-β1 medium (Univercell Biosolutions). DMEM containing 5.6 mmol/l glucose, 2% vol/vol Fetal Bovine Serum, 50 μmol/l 2-mercaptoethanol (Bio-Rad, Hercules, USA), 10 mmol/l nicotinamide (Calbiochem, Darmstadt, Germany), 5.5 μg/ml transferrin and 6.7 ng/ml selenite (Sigma-Aldrich) was used for transfection.

EndoC-βH1 cell line was Mycoplasma free as determined by the MycoAlert Mycoplasma Detection kit (Lonza). For the prevention of Mycoplasma contamination, Plasmocin Prophylactic (Invivogen, Toulouse, France) was added to the culture medium on a regular basis.

cDNA samples from human pancreatic islets were obtained from Cisanello University Hospital, Pisa, Italy. All the islets were isolated and cultured using the same experimental conditions and following established isolation procedures (15). Characteristics of islet preparations are described in Table S1 . The Ethical Committee of Cisanello University Hospital approved experiments using human islets.

LncTGM2 silencing in the EndoC-βH1 cell line was performed by transfecting 30 nmol/l of a siRNA targeting LncTGM2 (CD.Ri.214258.13.13, IDT) using Lipofectamine RNAimax reagent (Thermo Fisher Scientific) following the manufacturer’s instructions.

For overexpressing plasmids, LncTGM2 was purchased as a gBlock (IDT) and cloned into a modified pCMV6 vector using KpnI and FseI restriction enzymes (New England Biolabs, Ipswich, USA). Plasmids were transfected using Lipofectamine 2000 Transfection Reagent (Invitrogen, Carlsbad, USA) following the manufacturer’s instructions.

EndoC-βH1 cells were exposed to Actinomycin D (Sigma-Aldrich) at a final concentration of 5 μg/ml for 2, 4 or 6h. Palmitate treatment was performed by adding BSA-palmitic acid (0.5 mmol/l; 1:1) to DMEM/F-12, complemented with 0.25% vol/vol FBS, 50 μmol/l 2-mercaptoethanol (Bio-Rad), 10 mmol/l nicotinamide (Calbiochem), 5.5 μg/ml transferrin, 6.7 ng/ml selenite (Sigma-Aldrich), 100 units/ml penicillin and 100 μg/ml streptomycin (Lonza) for 4 or 8h.

For LncTGM2 RNA quantification in subcellular fractions of EndoC-βH1 cells, nuclei were isolated using C1 lysis buffer (1.28 mol/l sucrose, 40 mmol/l Tris -HCl pH 7.5, 20 mmol/l MgCl₂, 4% vol/vol Triton X-100). LncTGM2, MEG3 (nuclear control) and RPLP0 (cytoplasmic control) expression levels were measured by RT-qPCR and compared to the total amount of those RNAs in the whole cell lysate.

RNA extraction was performed using the NucleoSpin RNA Kit (Macherey Nagel, Düren Germany) and expression values were determined by RT-qPCR using iTaq Universal SYBR Green Supermix (Bio-Rad) using specific primers for each target RNA ( Table S2 ). All RT-qPCR measurements were performed in duplicates and expression levels were analyzed using the 2^–ΔΔCt method. A commercially available RNA panel set (Human total RNA master panel II, Clontech, Saint-Germain-en-Laye, France) was used to assess LncTGM2 and TGM2 expression levels in different human tissues.

EndoC-βH1 cells were washed with cold PBS and lysed in Laemmli buffer (62 mmol/l Tris-HCl, 100 mmol/l dithiothreitol (DTT), 10% vol/vol glycerol, 2% wt/vol SDS, 0.2 mg/ml bromophenol blue, 5% vol/vol 2-mercaptoethanol). Proteins in the lysate were separated by SDS-PAGE. After electrophoresis, proteins were transferred to nitrocellulose membranes using a Transblot-Turbo Transfer System (Bio-Rad) and blocked in 5% wt/vol non-fatty milk diluted in TBST (20 mmol/l Tris, 150 mmol/l NaCl and 0.1% vol/vol Tween 20) at room temperature for 1h. The membranes were incubated overnight at 4°C with a primary antibody specific for TGM2 (15100-1-AP, Proteintech Group, Rosemont, USA) diluted 1:1000 in 5% wt/vol BSA or anti-α-tubulin (Cat #T9026, Sigma-Aldrich) diluted 1:5000 in 5% wt/vol BSA. Immunoreactive bands were revealed using the Clarity Max Western ECL Substrate (Bio-Rad) after incubation with a horseradish peroxidase-conjugated anti-rabbit (1:1000 dilution in 5% wt/vol non-fatty milk) or anti-mouse (1:5000 dilution in 5% wt/vol non-fatty milk) secondary antibody for 1h at room temperature. The immunoreactive bands were detected using a Bio-Rad Molecular Imager ChemiDoc XRS and quantified using ImageLab software (Bio-Rad).

TGM2 promoter sequence was cloned into an empty pBV-Luc plasmid (Addgene, Watertown, USA) using KpnI and EcoRI restriction enzymes. EndoC-βH1 cells were transfected with a control vector (ovCTRL) or a vector overexpressing LncTGM2 (ovLncTGM2), and co-transfected with the TGM2 promoter reporter vector plus a pRL-CMV plasmid (used as an internal control) using Lipofectamine 2000 Transfection Reagent (Invitrogen). Dual-Luciferase Reporter Assay System (Promega, Madison, USA), was used to measure bioluminescence following the manufacturer’s protocol.

Secondary structure of LncTGM2 harboring the different alleles of rs2076380, rs7275079 and rs2067027 SNPs was predicted using the RNAsnp Web Server tool (16).

LncTGM2 harboring rs2076380-A or rs2076380-G alleles were in vitro transcribed using T7 RNA Polymerase kit (TaKaRa, Kusatsu, Japan). RNAs were run in a native TBE 2% wt/vol agarose gel and migration profile was analyzed in a ChemiDoc XRS apparatus (Bio-Rad).

For insulin release experiments, LncTGM2-silenced EndoC-βH1 were left in Opti-β2 (Univercell Biosolutions) starving medium for 24h. After glucose starvation, cells were incubated in KREBS medium (Univercell Biosolutions) for 1h, and consecutively exposed to 0 or 20 mmol/l glucose for 40 minutes. Supernatant and lysate were harvested and insulin release and content measured by a commercial human insulin ELISA kit (Mercodia, Uppsala, USA) according to the manufacturer’s instructions.

The association between the rs2076380 single variation in the TGM2 gene, clinical parameters and the risk of T2D was assessed using SPSS Statistics (IBM). Departures from Hardy-Weinberg equilibrium were tested in all groups using a chi-square goodness of fit test with one degree of freedom. The risk of developing T2D under exposure to rs2076380 TGM2 genotypes was evaluated using logistic regression to estimate Odd Ratios (OR), considering a dominant model in which G-allele carriers (i.e., AG-heterozygotes plus GG-homozygotes) were the reference group. To compare groups with respect to continuous variables, one-way ANOVA for multiple comparisons was used. Other statistical tests and plots were performed using GraphPad Prism 8 software (Dotmatics). Significance-level was set at p-value <0.05. Results for in vitro functional studies are represented as means ± standard error of mean (S.E.M.).

In order to determine the potential association of LncTGM2 with T2D clinical parameters, we performed an association study by genotyping a SNP located in the exonic region of LncTGM2 (rs2076380; chr20:38,165,027-38,165,227, hg38). This SNP can be considered as a tagSNP since it is in high linkage disequilibrium (LD>0.8) with other SNPs in the region ( Figure S1 ). The LncTGM2 SNP rs2076380 was tested in association with measures of T2D and other metabolic and clinical parameters in two independent cohorts ( Table S3 ). In cohort 1, the frequency of AA-individuals for the LncTGM2 SNP was 8.6%, similarly to the observed frequency in Cohort 2 (8.3%). These frequencies are in line with the observed frequency of the minor allele (A) in Caucasian populations (1000 Genomes Europe; A allele frequency = 0.32) (17) and Spanish control individuals (Medical Genome Project healthy controls from Spanish population; A allele frequency = 0.225) (18).

As observed in Figure 1 , the percentage of known type 2 diabetic individuals was increased in individuals harboring the rs2076380-AA genotype in both cohorts (Cohort 1: OR=1.13 [0.999-1.27], Pearson’s Chi-square p=0.006, two-sided Fisher’s exact test p=0.013); and Cohort 2: OR=1.08 [0.996-1.18], Pearson’s Chi-square p=0.018, two-sided Fisher’s exact test p=0.026). For both cohorts, regression analyses depicted the impact of the polymorphism in LncTGM2 on T2D incidence (ANOVA p-value of 0.026 in Cohort 1, and p=0.013 in Cohort 2) after correcting for sex and age. A codominant genetic model that included age, weight and sex effects was fitted to estimate the ORs between the exposure to the AA, AG and GG genotypes, the later as the reference group. The similar ORs for AG and GG genotypes obtained for the codominant model suggested the possibility of fitting a recessive model for AA-genotype carriers. This model allowed us to determine the OR between carriers of the AA genotype in relation to the G-allele porters. In this case, the residual deviance of the genotype, once age, weight and sex were added to the model, reached a p-value <0.05, indicating that the genotype effect was significant.

In addition, we observed that in Cohort 1, fasting glucose (p=0.005) and insulin levels (p=0.006) were increased compared to G-allele carriers ( Table S3 ). However, in Cohort 2, association with fasting glucose only reached statistical significance in female participants ( Table S3 ).

Previous studies have correlated LncTGM2 and TGM2 expression in tumor tissues and some human cell lines, including lymphoblast (K562), promyeoloblast (HL60) and monocyte (THP-1) cell lines (7). In order to clarify whether LncTGM2 and TGM2 expression was also correlated in healthy human tissues and in pancreatic beta cells, we first evaluated the expression of both genes in EndoC-βH1 cells and a set of human tissues. The highest expression of both, LncTGM2 and TGM2, was found in lung, placenta and heart, and the expression in the EndoC-βH1 cell line was similar to that of intestine and liver ( Figure S2 ). Spearman’s correlation analysis showed a significant correlation between LncTGM2 and TGM2 expression across the tissues analyzed (R=0.87 (0.59-0.9); p<0.0001). Interestingly, a correlation was also seen in EndoC-βH1 cells using siRNA-driven inhibition of LncTGM2. As shown in Figure 2A , a 70% decrease of LncTGM2 expression reduced TGM2 mRNA expression by 20%, suggesting a potential implication of LncTGM2 in the transcriptional regulation of TGM2.

In order to simulate the pathophysiological conditions of T2D in pancreatic beta cells, we next exposed EndoC-βH1 cells to palmitate (PA) as an in vitro model of lipotoxicity (19). As shown in Figure 2B , 4 and 8h PA exposure decreased both LncTGM2 and TGM2 expression in EndoC-βH1 cells, suggesting that in the presence of a lipotoxic insult the expression of both genes is reduced.

Knowledge of the subcellular localization of lncRNAs is crucial to understand and characterize their function. In contrast to protein-coding mRNAs, lncRNA themselves should be located in their site of action, and thus, their location within the cell is crucial for their function. While nuclear lncRNAs are usually implicated in the regulation of transcriptional activity, cytoplasmic lncRNAs can participate for example, in the regulation of mRNA stability or in protein translation (20). Having this in mind, we next decided to analyze the subcellular localization of LncTGM2 in EndoC-βH1 cells. As shown in Figure 2C , LncTGM2 was detected in both nuclear and cytoplasmic fractions, but its expression level was significantly higher in the nuclear compartment, suggesting its potential implication in transcriptional regulation. Since expression of LncTGM2 and TGM2 was significantly correlated in pancreatic beta cells, we performed a promoter reporter assay to clarify whether LncTGM2 was directly regulating the promoter activation of TGM2 gene. To this aim, we constructed an expression vector coding for a luciferase under the control of the promoter of TGM2. The luciferase vector was then co-transfected in EndoC-βH1 cells with an empty overexpression plasmid (ovCTRL) or with the overexpression plasmid of LncTGM2 (ovLncTGM2) and the activation of TGM2 promoter was determined by measuring bioluminiscence. As shown in Figure 2D , the activation of the TGM2 promoter was 1.5-fold higher in LncTGM2-overexpressing cells than in control cells, pointing out a role of LncTGM2 in the activation of TGM2 promoter, and consequently in the transcriptional activation of TGM2.

Disease-associated SNPs located within lncRNAs can affect their function through the disruption of their secondary structure (21, 22). As previously shown ( Figure S1 ), the T2D-associated rs2076380 SNP is in high LD with other two SNPs located in the exonic region of LncTGM2 (rs7275079 and rs2067027). To assess whether these SNPs alter the secondary structure of LncTGM2, we performed an in silico prediction analysis using the RNAsnp webserver from the Center for non-coding RNA in Technology and Health (23). Interestingly, rs2076380 was predicted to significantly alter the secondary structure of LncTGM2 (p=0.0803), while the software did not predict any significant change in the structure of the lncRNA when the different alleles of rs7275079 or rs2067027 SNPs were present (p>0.2) (data not shown). As shown in Figure 3A , the predicted secondary structures of LncTGM2 carrying the T2D protective (rs2076380-G) or risk allele (rs2076380-A) were significantly different. Consistent with the prediction, in vitro–transcribed forms of T2D protective and risk allele–harboring LncTGM2 revealed different motilities on a native agarose gel ( Figure 3B ), suggesting a different conformation of the lncRNA in the presence of one or other allele in rs2076380.

Taking into account that the secondary structure of a lncRNA is crucial for its interaction with other macromolecules, and thus, for its function (9), we next decided to determine whether the genotype of the T2D-associated SNP in LncTGM2 affected TGM2 expression in human pancreatic islets. To this aim we genotyped rs2076380 SNP and measured TGM2 expression in 16 cDNA samples from human islets, and performed an eQTL analysis. As shown in Figure S3 , there was a trend for higher expression of TGM2 in islets harboring the protective rs2076380-GG genotype compared to islets harboring the risk allele in heterozygosis (rs2076380-AG) or homozygosis (rs2076380-AA), although the differences did not reach statistical significance, probably due to the limited number of islets.

Next, to characterize the potential effect of each allele in rs2076380 SNP on the expression of both, LncTGM2 and TGM2, we constructed two LncTGM2 overexpression plasmids, one harboring the T2D risk allele (ovLncTGM2-A), and the other harboring the T2D protective allele (ovLncTGM2-G). Interestingly, allele-specific upregulation of LncTGM2 in beta cells revealed that the expression level reached by transfecting ovLncTGM2-G plasmid was higher than the expression level obtained with ovLncTGM2-A plasmid ( Figure 3C ), suggesting that the T2D risk allele might be affecting the stability of LncTGM2 RNA molecule.

In order to directly test whether the T2D-associated polymorphism affected LncTGM2 stability, we next performed an allele-specific overexpression of LncTGM2 and exposed the EndoCβ-H1 cells to Actinomycin D, a drug that inhibits transcription. As shown in Figure 3D , LncTGM2 harboring the protective allele (rs2076380-G) was more stable than the lncRNA harboring the risk allele (rs2076380-A) at all time-points, although the differences only reached statistical significance at 4h of Actinomycin D treatment (p<0.05). These results confirmed that the LncTGM2 risk allele in the T2D-associated rs2076380 SNP reduced the stability of the lncRNA.

To clarify whether the decreased stability of LncTGM2-A affected its capacity to regulate TGM2 expression, we next analyzed the expression of TGM2 in EndoC-βH1 cells overexpressing LncTGM2-A or LncTGM2-G. As observed in Figure 3C , only the upregulation of the lncRNA harboring the protective allele (ovLncTGM2-G) increased the expression of TGM2 mRNA. These results were also confirmed at the protein level ( Figure S4 ).

In summary, these results suggested that the LncTGM2 harboring the T2D risk allele induced less TGM2 expression due to its reduced stability.

Previous studies have shown that TGM2 might be implicated in insulin release through different mechanisms, including cytoplasmic actin remodeling and regulation of the action of other proteins during granule movement (24). Taking into account that our present results suggest that the T2D risk allele in LncTGM2 might induce a decrease in TGM2 expression in pancreatic beta cells, we next decided to determine the potential contribution of LncTGM2 in insulin release.

To this aim, we silenced LncTGM2 with a specific siRNA in EndoC-βH1 cells and determined glucose-stimulated insulin release. As shown in Figure 4 , high glucose stimulation in siCTRL-transfected EndoC-βH1 cells increased insulin secretion. In siLncTGM2-transfected beta cells, however, high glucose-induced insulin secretion (GSIS) was no longer statistically significant, suggesting that disruption of LncTGM2 in pancreatic beta cell might affect GSIS through diminished expression of TGM2.