Lysophosphatidylinositols Are Upregulated After Human β-Cell Loss and Potentiate Insulin Release

One of the hallmarks of both type 1 and type 2 diabetes (T2D) is the loss of functional β-cells. In T2D, β-cells attempt to compensate for initial peripheral insulin resistance by increasing the production of insulin (1–3). However, β-cells may deteriorate, and once a substantial fraction of functional β-cells has been lost, the insulin-producing cells can no longer cope with the increased demand and the clinical signs of diabetes appear (4,5).

Plasma metabolites can indicate the functional defects of individual tissues and their underlying cell types, such as the β-cells (6,7). For instance, we previously identified the deoxyhexose 1,5-anhydroglucitol as a reliable biomarker of potential changes in the β-cell mass in mouse models (8) and human cohorts (9). Noteworthy, deoxy- and dihydroceramides have been identified as potential T2D biomarkers (10) that may be elevated as early as 9 years before the onset of T2D (11). In addition, circulating triacylglycerols have been proposed as biomarkers of a crosstalk between the liver and the functional β-cell mass (7). Therefore, the identification of specific circulating metabolites that can warn of the loss of functional β-cells, before diabetes onset, are currently of utmost interest for more effective prevention of T2D. Furthermore, such metabolites that alter their levels in a state that does not yet qualify as diabetes could play a role in the maintenance of β-cell function and, therefore, contribute to the regulation of glucose homeostasis. In this study, using lipidomics analysis, we have identified lipid species associated with the loss of pancreatic β-cells triggering diabetes and that play a role in glucose homeostasis.

All animal experiments were conducted at the University of Geneva Medical Center with the approval of the animal care and experimentation authorities of the Canton of Geneva (approval GE/128/15). Mice were maintained on a 12-h dark/12-h light cycle with water and food ad libitum. All β-cell prohibitin 2 (β-Phb2^−/−) and control (Phb2^fl/fl) mice were generated and genotyped as previously described (8,12). The age of the mice is specified for each experiment. Upon sample collection, food was withdrawn from cages 3 h in advance to homogenize glycemia in an early fasting state.

We analyzed blood fraction samples stored at −80°C collected in the fasting state from participants who were already recruited for studies in Geneva, Switzerland; Rome, Italy; Maastricht, the Netherlands; and Amsterdam, the Netherlands.

Patients (n = 9: six women and three men; mean age 62.8 years ± 14.8 SD) undergoing pylorus-preserving pancreatoduodenectomy (PTX) were recruited from the Digestive Surgery Unit and studied at the Centre for Endocrine and Metabolic Diseases unit (both at the Agostino Gemelli University Hospital, Rome, Italy). Clinical characteristics are provided in Supplementary Table 1. The study protocol (ClinicalTrials.gov identifier NCT02175459) was approved by the local ethics committee (approval P/656/CE2010 and 22573/14), and all participants provided written informed consent, which was followed by a comprehensive medical evaluation.

Patients were scheduled for PTX, which was performed according to the pylorus-preserving technique (13). The volume of pancreas removed during the surgery is consistent (∼50%), as previously reported (14), maintaining almost the same remaining portion of the endocrine pancreas. This surgery is, therefore, a unique way to examine the effects of a sudden decrease of functional β-cell mass (9,15). The detailed inclusion and exclusion criteria for the patients were reported by Jiménez-Sánchez et al. (9).

A standard 75-g oral glucose tolerance test (OGTT) was performed by all participants before surgery and 40 (SD ±7) days after surgery (9). Matsuda indexes were calculated as indexes of whole-body insulin sensitivity based on insulin and glucose values obtained from the OGTT; β-cell function was evaluated by calculating the insulinogenic index as the change in insulin over the first 30 min divided by the change in glucose over the first 30 min (16). Integrated β-cell function was also measured, using the oral disposition index, calculated as the product of the insulinogenic index and the Matsuda index (17), which provides an assessment of insulin secretion in relation to insulin sensitivity.

Participants were recruited at Geneva University Hospital and they provided informed consent. The designed protocol was reviewed and approved by the institutional medical ethics committee (approval CER11–015). See the report of Hannich et al. (10) for further details on inclusion and exclusion criteria and details on blood sampling. Participants without antidiabetic treatments and with HbA_1c <6.0% (42 mmol/mol) were included in the control group of participants without type 2 diabetes and with normal glucose tolerance (NGT; n = 31). Participants taking an oral antidiabetic treatment or with HbA_1c ≥6.5% (48 mmol/mol) were included in the T2D group (n = 31). All the participants were asked to follow a moderate diet without excess fat or alcohol intake 24 h before the blood sampling day. All clinical data are reported in Supplementary Table 2, and diabetic medications are reported in Supplementary Table 3. HOMA indexes for these participants were calculated using the updated HOMA2 method. The disposition index was calculated as the product of the percentage of HOMA for insulin sensitivity (HOMA-S%) and the percentage of HOMA for β-cell function (HOMA-B%).

Patients at high risk of developing T2D participated in three different studies at Maastricht University: the Resveratrol-First-degree-Relatives (ReFdR) study (n = 12) (18), ACetylCarnitine in Trained (ACCT) study (n = 11) (19), and High Intensity Interval Training (HIIT) study (n = 7) (9). Their clinical characteristics are provided in Supplementary Table 4.

After an overnight fast, blood samples were drawn for measurements of plasma HbA_1c, plasma glucose, and serum insulin levels; and at time points 0, 90, and 120 min for measurement of serum insulin and plasma glucose levels for calculation of glucose clearance as defined by the oral glucose insulin sensitivity model (20). Inclusion and exclusion criteria for the ReFdR (18), ACCT (19), and HIIT (9) studies have been described previously. These patients were at increased risk for T2D (see Supplementary Methods for criteria) and stratified into two subgroups—HbA_1c ≥5.7% (n = 17) and HbA_1c<5.7% (n = 13)—according to the American Diabetes Association (ADA) criterion for prediabetes: HbA_1c >5.7% (21).

We included four individuals with overweight or obesity with T2D according to the 2010 ADA criteria: BMI, 25–40 kg/m²; male sex; aged 30–75 years; and using no glucose-lowering drugs other than metformin. Another four NGT control participants were included who had BMI ≤25 kg/m², were male, and were 30–75 years old. Their clinical characteristics are provided in Supplementary Table 5. The study was approved by the Institutional Review Board of the Amsterdam University Medical Centers (location Academic Medical Center) and performed according to the Declaration of Helsinki of October 2004. The study was registered with the Netherlands Trial Registry (no. NTR3234) and was performed between February 2012 and March 2013 at the Department of Endocrinology and Metabolism of the Amsterdam UMC. For a detailed description of the study protocol, please refer to the report of Stenvers et al. (22).

Lipid extracts were prepared using a modified methyl tert-butyl ether extraction protocol with addition of internal lipid standards, as previously described (10). Direct infusion tandem mass spectrometry (MS) analysis for the quantification of phospho- and sphingolipid species was performed using multiple reaction monitoring on a TSQ Vantage Extended Mass Range Mass Spectrometer (Thermo Fisher Scientific), equipped with a robotic nanoflow ion source (Triversa Nanomate; Advion Biosciences), as previously described (10). The quantification procedure was described by Pietiläinen et al. (23). Dried lipid extracts were resuspended in 250 µL of MS-grade chloroform and methanol (1:1) and further diluted in either chloroform and methanol (1:2) plus 5 mmol/L ammonium acetate (negative ion mode) or in a mix of chloroform, methanol, and H₂O (2:7:1) plus 5 mmol/L ammonium acetate (positive ion mode). Lipid concentrations were calculated relative to the relevant internal standards and then normalized to the total phosphate content of each total lipid extract and expressed as a percentage of total lipids detected. Lipid concentrations were not corrected for class II isotopic overlaps. Unless specified, multiple paired t tests were performed for the Rome cohort and unpaired for all the rest of cohorts and mouse experiments (24). P values and false discovery rates were computed. Lipids were considered significant when P < 0.05. Hierarchical cluster analysis and data visualization as heatmaps were based on the Ward algorithm with Euclidean distance using MetaboAnalyst 5.0 software (Quebec, Canada).

Mouse pancreatic islets were isolated through collagenase digestion and cultured overnight in complete RPMI-1640 medium, as described previously (25). Human islets were isolated from pancreases of deceased multiorgan donors (NGT group: n = 3; T2D group: n = 1). Islets from the donors without diabetes, who provided informed consent, were provided by the European Consortium for Islet Transplantation, and the islets from the T2D group were purchased from Tebubio Srl. (Paris, France). Donor information is shown in Supplementary Table 6.

Islets were maintained for a standard recovery period (between 1 and 4 days) in CMRL-1066 medium containing 5.5 mmol/L glucose supplemented with 10% (v/v) FCS and antibiotics. For secretion experiments, batches of 25 islets were handpicked and preincubated with Krebs-Ringer bicarbonate HEPES buffer supplemented with 2.8 mmol/L glucose and 0.1% BSA at 37°C for 1 h. Subsequently, islets were incubated with basal glucose (2.8 mmol/L) with a commercial mixture of bovine liver lysophosphatidylinositols (lysoPIs) containing 90% lysoPI18:0, 5% lysoPI18:1, and 5% lysoPI16:0 (Avanti Lipids) at the specified concentration dissolved in DMSO or vehicle and, subsequently, at the specified stimulatory glucose concentration for 1 h. At the end of the incubation period, supernatants were collected, and islets were resuspended in Trizol. Insulin in supernatants was measured using a radioimmunoassay (Millipore) or human insulin ELISA kit (Mercodia, Uppsala, Sweden).

Total RNA was extracted from isolated islets from prediabetic (β-Phb2^−/−) and control (β-Phb2^fl/fl) mice with the Trizol reagent (Invitrogen, Carlsbad, CA) and 2 µg was converted into cDNA, as described previously (26). A list of primers is given in Supplementary Table 7. RT-qPCR was performed using an StepOnePlus Real-Time PCR System (Thermo Fisher Scientific, Waltham, MA). PCR products were quantified fluorometrically using the SYBR Green Master kit (Roche, Mannheim, Germany). Experiments were performed in duplicate for the gene Gpr119, and mean values were normalized to those of the reference mRNA ubiquitin C (Ubc).

Total RNA was extracted from cultured isolated human islets with Trizol reagent (Invitrogen) (26). RNA sequencing was performed in Susanne Mandrup’s laboratory (University of Southern Denmark, Odense, Denmark), as detailed previously (27). Results are expressed in log2-fold changes (log2FCs), and individual adjusted P values were calculated for each donor. Significant changes were considered when two or more independent islet batches (donors) exhibited down or upregulation with a log2FC threshold of 0.5 associated with at least one adjusted P < 0.05. For network analysis of RNA-sequencing data, see Jiménez-Sánchez et al. (27).

Modified INS-1E β-cells (Research Resource Identifier: CVCL_0351) expressing a Gaussia luciferase in place of the insulin c-peptide (28,29) were used for luminescence-based secretion assay. Cells were cultured in RPMI-1640 GlutaMAX medium at 11.1 mmol/L glucose supplemented with 10 mmol/L HEPES, 5% (v/v) heat-inactivated FCS, 100 U/mL penicillin, 100 μg/mL streptomycin, 1 mmol/L sodium pyruvate, and 50 μmol/L β-mercaptoethanol. For secretion experiments, cells were starved for 2 h in glucose-free RPMI-1640 medium, washed, and incubated for another hour with basal glucose (2.8 mmol/L) in Krebs-Ringer bicarbonate HEPES buffer with lysoPIs mix 40 μmol/L dissolved in DMSO or vehicle. We added 5 μmol/L native coelenterazine (Nanolight Technologies, Pinetop, AZ) to the wells, incubated that for 20 min, and luminescence emitted upon 7 and 15 mmol/L glucose online stimulation was monitored for 65 min using the FLUOstar microplate reader (26).

Data and materials from this study are available upon reasonable request to one of the corresponding authors.

To investigate the circulating lipid signature associated with loss of functional β-cells, we measured lipid metabolites from 1) prediabetic β-Phb2^−/− mice, 2) patients after PTX resulting in acute β-cell loss, 3) patients with established T2D, and 4) patients at high risk of developing diabetes, by lipidomic analyses.

β-Phb2^−/− mice exhibit a spontaneous and progressive loss of β-cells, ultimately leading to diabetes from the age of 6 weeks onward (8,12). The prediabetic stage (4–6 weeks of age) is characterized by a gradual decline in β-cell mass, as documented previously (8,12), although this is essentially asymptomatic (Supplementary Fig. 1). At 5 weeks of age, when euglycemic prediabetic β-Phb2^−/− mice have lost about half of their β-cells (8), we observed an early alteration of circulating lipid metabolites. Using lipidomics analysis, we identified a total of 340 circulating lipid metabolites (Fig. 1A) across the following major lipid classes: phosphatidylethanolamines (PEs), phosphatidylcholines (PCs), phosphatidylinositols (PIs), ceramides (Cers), and hexosylceramides (HexCers) (see Supplementary Table 8). Overall, PC metabolites (diacylPCs, ether [O-] PCs, and lysoPCs), together with diacylPIs, were the most abundant lipid classes (Fig. 1B, Supplementary Table 11) detected in β-Phb2^−/− mouse plasma. A heatmap of the top 25 regulated lipids using hierarchical clustering shows differentiation between the prediabetic and control mice (Fig. 1D). Indeed, lysoPIs along with other PI metabolites, were increased in these prediabetic mice, together with decreased diacylPC, Cers, PE O-, and diacylPI species (Fig. 1E). LysoPIs constituted the top and only significantly regulated lipid class among all lipid classes detected (Fig. 1C and 1F, Supplementary Fig. 2). These changes in the lysoPI family were essentially contributed by lysoPI14:1, lysoPI14:2, lysoPI18:0, and the most abundant, lysoPI22:0 (Fig. 1G), which were among the top 25 regulated lipids (Fig. 1D). Of note, these lysoPIs were not altered in the liver, and we observed a trend for decreased lysoPI22:0 in the muscle of these prediabetic mice (Supplementary Fig. 2).

We next analyzed the plasma lipid profile of patients after an acute loss of their β-cells. These patients without known diagnosis of diabetes underwent PTX (Rome cohort; n = 9) (Supplementary Table 1). All patients were evaluated 1 week before surgery and 40 (SD ±7) days (range, 34–48 days) after surgery. The volume of pancreas removed during the surgery was constant (∼50%), as previously reported (14). Thus, this surgery represents a model of acute ∼50% reduction of the β-cell mass (15,30). Patients were divided according to their glucose tolerance status as determined by an OGTT performed 1 week before the surgical procedure. According to the ADA classification (31), patients whose 2-h postload glucose was <140 mg/dL were referred to as NGT (n = 2), those whose 2-h postload glucose was 140–199 mg/dL were defined as impaired glucose tolerant (IGT; n = 6), and one patient with 2-h postload glucose >200 mg/dL was classified as havingT2D.

As expected, PTX resulted in lower insulin secretory capacity, revealed by the patients’ insulinogenic index and lower insulin levels during OGTT (Supplementary Table 1), with two patients remaining NGT, five patients IGT, and two patients T2D. We then compared the plasma lipid profiles assessed by lipidomics analysis before and after PTX. We identified a total of 433 circulating lipid metabolites, including PEs, PCs, PIs, Cers, HexCers, and cardiolipins (CLs) (Supplementary Table 9, Fig. 2A). The composition of the human circulating lipid profile of these patients was comparable to that of mice, with PC metabolites (diacylPCs, O- PCs, and lysoPCs), together with diacylPIs, being the most abundant lipid classes (Fig. 2B, Supplementary Table 11). After removal of half of the pancreas, patients had altered individual lipid metabolites, as shown on the heatmap of the top 50 regulated lipids upon hierarchical clustering (Fig. 2D). Specifically, there was an increase in sphingomyelins (SMs), Cers, HexCers, CLs, PIs (diacylPIs, O-PIs, and lysoPIs), lysoPEs, and diacylPEs, whereas a decrease was observed for lysoPCs and changes in Cers, PC(O-)s, and diacylPCs single-lipid species (Fig. 2E), showing distinct separation between the PTX versus the basal state. When looking at the overall amount of lipid metabolites per lipid class, patients had increased SM, HexCer, CL, PI metabolites (diacylPIs, O-PIs, and lysoPIs), diacylPEs, and decreased PCs (Supplementary Fig. 3), with lysoPIs being the top regulated lipid class (Fig. 2C and 2 F). After the hemipancreatectomy, 14 of the 15 lysoPI individual species detected were increased (Fig. 2G), with lysoPI 22:1, 16:0, and 16:1 being among the top 50 regulated lipids (Fig. 2D). Of note, lysoPI levels were increased regardless of their glucose control status and we found no correlation with glycemic control parameters in this cohort (data not shown). All in all, the changes in lysoPI levels observed in prediabetic mice with partial β-cell loss were confirmed in this cohort of patients after removal of 50% of their pancreas.

Next, circulating levels of lysoPIs were analyzed in cohorts of patients with established T2D (n = 31) and in participants with NGT (n = 31) (see Supplementary Table 2). We identified a total of 307 circulating lipid metabolites across several lipid classes (Supplementary Table 10, Fig. 3A). Like the hemipancreatectomized human cohort and the prediabetic mouse model, the diacyl-, and O-PCs, along with diacylPIs, were the most abundant lipid classes among the circulating serum lipids (Fig. 3B, Supplementary Table 11). A hierarchical clustering analysis based on the top 50 regulated lipids revealed two distinct clusters corresponding mainly to patients with T2D rather than participants with NGT, with some interindividual differences accounting for an imperfect classification among these two groups (Fig. 3D). Patients with T2D had modified individual lipid metabolites, with increased Cer and PI metabolites, reduced PC metabolites and PE (O-)s, and altered diacylPEs (Fig. 3E). Again, we observed that lysoPIs were among the top regulated lipid classes (Fig. 3C and 3 F), with lysoPI 18:1, 18:2, 20:0, 20:1, 20:2, and 22:0 being significantly upregulated (P < 0.05), and lysoPI 18:1, 18:2, and 20:0 being among the top 25 regulated lipids in the patients with T2D versus participants with NGT. Our recently published study (32) characterized the lipidomic temporal profile of human pancreatic islets derived from 10 participants without diabetes and 6 donors who had T2D, using the same MS-based lipidomic analyses. Overall, lysoPIs levels extracted from our previously published data set did not differ between pancreatic islets derived from donors with T2D and their NGT counterparts (Fig. 3H), with lysoPI14:0 being significantly reduced in islets of donors with T2D (Fig. 3I). These results suggest that in T2D, pancreatic islets do not accumulate the increased circulating lysoPIs. Concomitantly, we observed increased expression of the enzyme that esterifies arachidonyl-coenzyme A to lysoPI to regenerate PI (membrane-bound O-acyltransferase domain-containing protein 7 [MBOAT7]) in isolated human islets exposed to diabetogenic conditions (Supplementary Fig. 3).

We next explored associations between circulating lysoPIs and the different metabolic parameters determined in the participants. Although clear correlations among circulating lysoPIs and the soluble biomarker of functional β-cell mass 1,5-anhydroglucitol (9), fasting glycemia, HbA_1c, and disposition index of static measurements (HOMA B% × HOMA S%) were observed, there was no correlation with fasting circulating insulin (not shown), HOMA for insulin resistance, or BMI (Fig. 4). This may point to a link of circulating lysoPIs with β-cell function rather than with insulin resistance. Interestingly, circulating lysoPI levels were comparable between patients requiring oral antidiabetic drugs as compared with patients requiring insulin therapy, suggesting disconnection with glycemic control per se or circulating insulin (Supplementary Fig. 5). Consistent with these observations, additional collection of serum from an independent small cohort of patients with T2D and NGT control participants subjected to lipidomics analyses revealed an increase of lysoPI 18:0 in serum of patients with T2D (Supplementary Fig. 6A).

Finally, we analyzed a cohort of patients identified as being at high risk for developing T2D (i.e., having altered glucose homeostasis [see definition criteria in Materials and Methods]; full data not shown), with HbA_1c of 5.7% set as a threshold for separation between two subgroups according to the ADA criterion for prediabetes (21). We measured increased lysoPI 18:0 in the sera of the selected study participants with HbA_1c >5.7%, who thus were at high risk for T2D (Supplementary Fig. 6B). In summary, lysoPI species were upregulated in the diabetic and prediabetic state, correlating with indicators of β-cell mass and function.

Before the onset of T2D, β-cells may increase the release of insulin to compensate for insulin resistance. Because circulating lysoPIs were increased in the diabetic but also in the prediabetic state, we explored their potential role in regulating the function of β-cells. LysoPIs stimulate insulin secretion in pancreatic islets through calcium mobilization (33). Some studies have highlighted GPR55 as the potential receptor for the signaling effects of lysoPIs in several cell types. In islets, however, this signaling pathway seems to be independent of GPR55 (33). Some reports have shown that GPR119 is expressed in β-cells and is activated by lysoPC, which is structurally similar to lysoPI, resulting in the stimulation of insulin secretion (34). This points to GPR119 as a potential receptor mediating an insulinotropic effect of lysoPIs in pancreatic islets. We observed that GPR119 is expressed in pancreatic islets isolated from control and prediabetic mice, as well as in human islets (Fig. 5A and B). Human islets exposed to glucotoxic conditions in vitro (25 mmol/L glucose for 3 days) exhibited downregulation of GPR119, as shown by RNA sequencing (full data not shown; donors 1–5) (Fig. 5B). Indeed, lysoPI facilitates COPII vesicle budding from the endoplasmic reticulum, probably via their biophysical properties (35), as demonstrated in yeast. Therefore, we next explored the potential impact of lysoPIs on insulin secretion using INS-1E β-cells expressing G-luciferase in place of the bioinactive insulin c-peptide (28) allowing real-time monitoring of secretion (26).

In humans, plasma levels of lysoPIs are in the micromolar range and higher (36), which prompted us to use 40 μmol/L for in vitro studies. INS-1E β-cells exposed to 40 μmol/L lysoPIs (90% lysoPi18:0, 5% lysoPI18:1, 5% lysoPI16:0) were incubated at 2.8 mmol/L glucose (G2.8; basal) for 15 min before stimulation with sequential stimulatory glucose concentrations of 7.5 mmol/L and 15.0 mmol/L (G15). Upon glucose stimulation, lysoPI treatment induced a strong potentiation of both the first and the second phases of insulin secretion (Fig. 5C and D). Of note, the responses to a depolarizing concentration of KCl were similar in lysoPI-treated as compared with untreated INS-1E cells (data not shown), suggesting no additive effects of lysoPIs on membrane potential.

Next, we confirmed the effects of lysoPIs on insulin secretion in isolated mouse islets. Acute treatment (15 min) with 40 μmol/L lysoPI mix (90% lysoPi18:0, 5% lysoPI18:1, 5% lysoPI16:0) potentiated glucose-stimulated insulin secretion at 11 mmol/L (+256%; P < 0.0001), 15 mmol/L (G15; +231%; P < 0.0001), and 25 mmol/L glucose (G25) (+230%; P < 0.0001) (see Fig. 6A). The effect of lysoPIs was dose dependent (Supplementary Fig. 7).

We then tested if lysoPIs could potentiate glucose-stimulated insulin secretion in islets isolated from β-Phb2^−/− mice at different stages of diabetes progression (prediabetic and diabetic states). Addition of lysoPIs strongly potentiated glucose-stimulated insulin secretion at G15 in islets from prediabetic (+332%; P = 0.06) as well as diabetic (+342%; P < 0.0001) mice (Fig. 6B). Finally, we confirmed the insulinotropic effect of lysoPIs in isolated islets from three human donors without diabetes (+67%, P = 0.0003; +569%, P < 0.0001; +434%, P < 0.0001, respectively). We also observed such an effect on islets isolated from a T2D donor (+216%; P < 0.0001; Fig. 6C).

Overall, lysoPIs potentiated glucose-stimulated insulin secretion in INS-1E β-cells and in mouse and human islets. Importantly, lysoPIs partially rescued the impaired secretory response of mouse and human islets isolated from individuals with prediabetes and those with diabetes.

In this study, using in-depth lipidomics analyses, we have identified circulating lipid species associated with the loss of pancreatic β-cells eventually leading to diabetes. We report that higher lysoPIs represent the major plasma lipid signature in prediabetic mice that are associated with lower β-cell mass at an asymptomatic stage before diabetes onset.

A similar increase in plasma lysoPIs was observed in a cohort of patients with varying glucose tolerance status who underwent PTX that led to acute reduction of approximately 50% of β-cell mass. It was previously reported that circulating levels of lysoPIs are doubled in individuals with obesity and T2D (37). Taken together, these results indicate that lysoPI levels increase in mice and humans along with a decline in functional β-cell mass before clinical symptoms of diabetes occur. Interestingly, lipidomic characterization of a genetically modified mouse model of early stages of human type 1 diabetes showed an increase in lysoPIs in the early stages of type 1 diabetes pathogenesis (38). Collectively, these findings are supportive for a link between elevated lysoPIs and loss of functional β-cell mass, preceding clinical signs of diabetes. Although not yet associated with the β-cell mass, circulating lysoPIs have previously been linked to an increased risk of developing T2D in individuals with otherwise normal baseline HbA_1c levels (39). Indeed, elevated serum lysoPI levels have been reported before the onset of diabetes (40,41).

Furthermore, we found that patients with established T2D had higher circulating lysoPI levels. Elevated lysoPI levels correlated significantly with HbA_1c (reflecting glycemic control), fasting glycemia, and disposition index, but not with insulin resistance or obesity. These results are consistent with the observed elevated levels of plasma lysoPIs in the diabetic db/db mice (42). That same study, by Kurano et al., found higher serum lysoPI levels in patients with T2D, but only in the lipoprotein-depleted fractions, as opposed observations in db/db mice (42).

Because lipoproteins are produced exclusively by the enterocytes in the intestine and the hepatocytes in the liver, more research is needed to identify the circulating carriers of the lysoPI fraction influenced by T2D. Accordingly, the tissue source of lysoPIs, or the pathways of its impaired clearance, should be instrumental for our understanding of the mechanisms linking its plasma levels and the functional β-cell mass. Moreover, in vivo administration of lysoPIs, not tested in the present study, should help determine if lysoPI’s natural endogenous elevation upon β-cell loss is sufficient to reach optimal effects or if increase by exogenous supply could be beneficial. Our results show that the lysoPI species upregulated in the human cohorts and the mouse model correspond to the most abundant lysoPIs, suggesting higher turnover and activity of the enzyme phospholipase A2, thereby increasing the production of major lysoPIs (43,44). More studies are required to elucidate the metabolic effects of the different acyl chain lengths and saturation degree of the lysoPIs.

Importantly, we observed that lysoPIs affected β-cell functionality by increasing glucose-stimulated insulin secretion. This effect was previously attributed to the ability of lysoPIs to induce intracellular calcium mobilization in rat and mouse pancreatic islets (45,46). However, it is unknown whether this effect of lysoPIs is mediated by receptors or by receptor-independent mechanisms. GPR55, which is expressed in the brain, intestine, and endocrine pancreas, has been proposed as a potential lysoPI receptor (39,44,47). The role of GPR55 in pancreatic β-cells is debatable. The GPR55 agonist O-1602 increased glucose-stimulated insulin secretion in rodent islets in a GPR55-dependent way (48), and others observed that the lysoPI effects on the secretory response are preserved in islets isolated from GPR55-null mice (33). Interestingly, it has been shown that the GPR119 receptor is expressed in the β-cell, and that activation by lysoPC, which is structurally similar to lysoPI, potentiates glucose-dependent insulin secretion (34). GPR119 has also been implicated in the lysoPI response of enteroendocrine cells by potentiating GLP1 secretion independently of GPR55 (49,50). In the present work, we did not detect expression of GPR55 in human islets using RNA-sequencing data, although we observed downregulation of GPR119 in human islets upon chronic metabolic stress in vitro, pointing to GPR119 as the likely candidate receptor for lysoPIs. LysoPIs could also exert receptor-independent effects.

For instance, it has been demonstrated that lysoPIs activate the two-pore potassium channel TREK subunit and transient receptor potential cation channel subfamily V member 2 channels in different cell types (51,52). These channels modulate cytosolic calcium concentrations, thereby influencing insulin exocytosis (53,54). Furthermore, lysoPIs may act more directly on the secretory pathway; they have been shown to facilitate COPII budding from the endoplasmic reticulum (35).

Our study provides unprecedented evidence for a beneficial role of lysoPIs on islets isolated from mice and humans with prediabetes and T2D, respectively. Indeed, lysoPIs significantly rescued impaired glucose-stimulated insulin secretion in mouse islets with varying degrees of β-cell loss, and, in human islets derived from a patient with T2D, suggesting a compensatory mechanism that is activated in the context of diabetes.

In this study, we used an unbiased lipidomics strategy to identify new lipid species associated with the loss of pancreatic β-cells triggering diabetes. We found higher lysoPIs were the major circulating lipids in prediabetic mice with controlled and progressive β-cell loss, a pattern confirmed both in a cohort of patients who underwent acute removal of 50% of their pancreas and in patients with T2D. Increased lysoPIs significantly correlated with HbA_1c levels, which reflect glycemic control; with fasting glycemia; and disposition index and did not exhibit a correlation with insulin resistance or obesity. Addition of lysoPIs to INS-1E β-cells and isolated mouse and human pancreatic islets enhanced glucose-stimulated insulin secretion, even when islets displayed secretory defects associated with diabetes. Overall, circulating lysoPIs levels increase upon β-cell loss and exert concomitant insulinotropic effects on β-cells.

Acknowledgments. The authors thank Jacques Philippe and Emmanouil Dermitzakis for invaluable contribution to the Geneva cohort study, Giovanni D’Angelo for the help with the lipidomics studies, and Marie-Claude Brulhart-Meynet for the technical assistance with the lipidomics experiments. The authors are deeply saddened by the tragic death of Steven A. Brown, our most valued colleague and dear friend, who made essential contributions to the Geneva cohort study underlying this work.

Funding. This work was supported by Swiss National Science Foundation grants 310030_184708/1 (to C.D.) and 310030_192486 (to P.M.); the Vontobel Foundation; the Novartis Consumer Health Foundation; EFSD/Novo Nordisk Program for Diabetes Research in Europe; the Olga Mayenfisch Foundation; Fondation pour l'innovation sur le cancer et la biologie; Ligue Pulmonaire Genevoise, Swiss Cancer League (grant KFS-5266-02-2021-R); Velux Foundation; Leenaards Foundation; Swiss Life Foundation; the ISREC Foundation; the Gertrude von Meissner Foundation (to C.D.); the Bo & Kerstin Hjelt Diabetes Foundation and Fundación Alfonso Martín Escudero (to C.J.S.); by Swiss Life Foundation and Young Independent Investigator Grant from the Schweizerischen Gesellschaft für Endokrinologie und Diabetologie (SGED) (to F.S.); by grants from the Università Cattolica del Sacro Cuore (Fondi Ateneo Linea D.1, 2020, and Fondi Ateneo Linea D.1, 2021); and by the Italian Ministry of Education, University and Research (grant GR-2018-12365577 to T.M.). Human islets were provided by the Geneva University Hospital through JDRF award 31-2008-416 (European Consortium for Islet Transplantation Islet for Basic Research program; Swiss Ethics Committee, Commission Cantonale d’Ethique de la Recherche sur l’être humain de Genève [CCER] protocol 2016-01979 and 2017-00315).

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

Author Contributions. C.J.-S., C.D., and P.M. planned and conducted the study. C.J.-S., F.S., T.M., U.L.-M., L.L., J.P.M., G.D.G., G.Q., I.G., F.J., P.S., D.J.S., S.A., A.G., H.R., C.D., and P.M. contributed to the production and analysis of data. E.B., P.C., and D.B. provided human islets. C.J.-S., C.D., and P.M. drafted the manuscript. All authors edited the manuscript and approved the final version. C.D. and P.M. are the guarantors of this work and, as such, had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.

Prior Presentation. This work was presented in part at the Swiss Lipid Workshop, Bern, Switzerland, September 2022.