Circulating long non-coding RNAs as predictors of type 2 diabetes mellitus development: results from the CORDIOPREV study Authors: Hatim Boughanem, María José Párraga-Viúdez, Ana Ojeda-Rodríguez, Juan Luis Romero-Cabrera, Eugenia Ruiz-Díaz Narváez, Alejandro López-Moreno, José David Torres-Peña, Manuel Macías-González, Juan Francisco Alcalá-Díaz, Antonio Pablo Arenas-de Larriva, Pablo Pérez-Martínez, Javier Delgado-Lista, José López-Miranda, Oriol Alberto Rangel-Zuñiga Published: 2025 ABSTRACT AbstractBackgroundType 2 diabetes mellitus (T2DM) is a growing global health challenge. Conventional diagnostic tools have limited sensitivity and specificity for early-stage disease. In this context, long non-coding RNAs (lncRNAs) have emerged as promising biomarkers for T2DM risk. However, studies exploring their predictive value remain limited. This study aimed to evaluate the potential of circulating lncRNAs in T2DM development and to assess their interaction with dietary interventions.MethodsThe study included 462 non-diabetic participants from the CORDIOPREV study, followed for 5 years under Mediterranean or low-fat diet interventions. Expression levels of 22 circulating lncRNAs were quantified by qPCR. A series of statistical and machine learning methods were applied.ResultsRandom forest analysis identified four lncRNAs (XIST, LINC01116, CASC2, LINC01370) as predictive of T2DM incidence. The combination of these lncRNAs with clinical variables significantly improved prediction performance (AUC = 0.730) compared to models with Hb1Ac (p = 0.015) or clinical variables alone (p < 0.001). Regarding the constructed lncRNA score from the previous model, a lower lncRNA score was associated with a reduced risk of developing T2DM (HR 0.37 (0.24–0.59), p < 0.001), and showed an inverse correlation with the disposition index among individuals following a Mediterranean diet (r = − 0.18, p = 0.009).ConclusionCirculating lncRNAs, particularly integrated into an epigenetic score, represent promising predictive biomarkers for T2DM development. The interaction with dietary intervention—especially the Mediterranean diet—supports their potential use in guiding personalized dietary strategies for T2DM prevention in high-risk populations.Supplementary InformationThe online version contains supplementary material available at 10.1186/s12933-025-02942-y.Keywords: LncRNAs, T2DM, Mediterranean diet, Epigenetic, CORDIOPREV Introduction Type 2 diabetes mellitus (T2DM) currently represents one of the major threats to global public health, with its prevalence expected to rise further in the coming years [1]. According to recent projections by the International Diabetes Federation (IDF), more than 852 million individuals are expected to be living with T2DM by 2050, potentially increasing its global prevalence to approximately 45% [2]. Despite these alarming figures, a substantial proportion of cases remains undiagnosed, delaying the initiation of treatment and further increasing the risk of developing diabetes-related complications [1, 2]. The conventional diagnosis of T2DM relies on biochemical parameters such as fasting plasma glucose, glycated hemoglobin (HbA1c), and the oral glucose tolerance test (OGTT) [3]. However, these methods may fail to detect the early stages of T2DM. Furthermore, commonly used predictive tools—such as FINDRISC [4] and ADA risk score [5]—may oversimplify the complexity of T2DM. These limitations highlight the urgent need to identify novel biomarkers capable of increasing the predictive capacity for T2DM incidence. Considerable efforts are underway to identify potential predictive biomarkers for T2DM, with particular focus on epigenetic biomarkers. Among emerging candidates, long non-coding RNAs (lncRNAs) have garnered increasing attention due to their regulatory roles in various metabolic and pathophysiological processes associated with T2DM [6]. LncRNAs have been shown to play key roles in metabolic homeostasis, including glucose and lipid metabolism, pancreatic β-cell function, cell proliferation, insulin resistance and chronic inflammation [6–9]. Additionally, lncRNAs are stable in circulation and can be readily detected in bloodstream [10], supporting their potential as biomarkers. For instance, Yundung et al. [11] described a panel of differentially expressed lncRNAs in smooth muscle cells from patients with T2DM and peripheral artery disease, while Wenzl et al. [12] demonstrated that certain circulating lncRNAs were prospectively associated with future onset of T2DM in older adults. However, further large-scale, longitudinal, and clinically validated studies are needed to confirm their reliability as predictive biomarkers. In this context, the CORDIOPREV study offers a valuable framework to investigate these mechanisms, as individuals with previous cardiovascular disease have an increased risk of developing T2DM, thereby increasing the risk of recurrent events [13]. Therefore, this study aimed to evaluate the potential of lncRNAs as epigenetic biomarkers for predicting T2DM development in patients with coronary heart disease. Additionally, the interaction between dietary interventions (Mediterranean or low-fat diets) and lncRNA expression was assessed to determine whether these biomarkers could predict the effectiveness of dietary strategies in preventing T2DM and guide the selection of the most appropriate diet to reduce T2DM risk. Materials and methods Study design This study was conducted within the framework of the CORDIOPREV study (https://clinicaltrials.gov, ID: NTC00924937), a randomized, controlled, prospective, single-blinded dietary intervention trial designed to evaluate the long-term effects of two dietary patterns on secondary prevention in patients with cardiovascular disease. The study included 1002 participants aged 20–75 years, all of whom were at high cardiovascular risk and had experienced a coronary event at least six months prior to inclusion. Participants were recruited between October 2009 and February 2012, primarily at the “Reina Sofía” University Hospital (Córdoba, Spain). In addition to standard medical therapy for cardiovascular disease, participants were randomly assigned to follow either a Mediterranean diet or a low-fat diet. The intervention had a median follow-up period of 7 years and the rationale, methodology, and baseline characteristics of the participants have been previously described [14, 15]. The study protocol and its amendments were approved by the Ethics Committee of “Reina Sofía” University Hospital and adhered to the principles of the Declaration of Helsinki and Good Clinical Practice. All participants provided written informed consent prior to enrollment. Study population In the present work, all the subjects (n = 462) who had not been clinically diagnosed with T2DM and were not receiving pharmacological treatment for glycemic control, at baseline in the CORDIOPREV study were included. The criteria for the diagnosis of T2DM were those established by the American Diabetes Association (ADA) [16], which include fasting plasma glucose (FPG) ≥ 126 mg/dL, or 2-h plasma glucose ≥ 200 mg/dL during an OGTT, or HbA1c levels ≥ 6.5%. After a median follow-up of 60 months under dietary intervention, and according to the ADA criteria, 107 participants developed T2DM (Incident-T2DM) (42 following a low-fat diet and 65 a Mediterranean diet), while 355 remained non-diabetic (Non-Incident) (174 following a low-fat diet and 181 a Mediterranean diet). Biochemical determinations Fasting venous blood samples were collected after an overnight fast of at least 12 h and processed by centrifugation at 7000 × g for 15 min at 4 °C to obtain serum. Serum concentrations of total cholesterol (TC), triglycerides (TG), and high-density lipoprotein cholesterol (HDL-c) were analyzed using automated enzymatic colorimetric methods [19, 20] on a modular DDPPII Hitachi autoanalyzer (Roche Diagnostics, Basel, Switzerland), with specific reagents supplied by Boehringer-Mannheim (Germany). Low-density lipoprotein cholesterol (LDL-c) was estimated using the Friedewald formula [17]. Fasting plasma glucose was quantified using the hexokinase method. Plasma insulin concentrations were measured using a microparticle enzyme immunoassay (Abbott Diagnostics, Matsudo-shi, Japan). Based on glucose and insulin values obtained during the OGTT, several metabolic indices were calculated to assess insulin sensitivity and β-cell function. These included the Homeostasis Model Assessment of Insulin Resistance (HOMA-IR), the Insulinogenic Index (IGI), and the disposition index (DI) following previously described methodologies [21, 22]. The FINDRISC score was calculated as defined previously [4, 5, 17]. Dietary assessment As part of the CORDIOPREV study, participants were randomly assigned to follow one of two dietary patterns: either a low-fat, high-complex carbohydrate diet recommended by the National Cholesterol Education Program and the American Heart Association, comprising < 30% total fat [< 10% saturated fat, 12–14% monounsaturated fatty acids (MUFA), and 6–8% polyunsaturated fatty acids (PUFA)], 15% protein, and a minimum of 55% carbohydrates—or a Mediterranean diet rich in olive oil, with 35% of calories from fat (22% monounsaturated, 6% polyunsaturated, < 10% saturated), and a maximum of 50% carbohydrates. A detailed description of dietary assessment methods and follow-up visits have been previously published [15, 18–20]. RNA isolation, cDNA synthesis and qPCR Total RNA was isolated from 200 μL of plasma using the RNeasy Mini Kit (QIAGEN, Hilden, Germany), following the manufacturer’s instructions. The aqueous phase obtained after centrifugation was processed using silica membrane spin columns, and RNA was eluted in 14 μL of RNase-free water and stored at − 80 °C until analysis [17]. Complementary DNA (cDNA) synthesis and simultaneous preamplification were performed using the iScript™ Explore One-Step RT and PreAmp Kit (Bio-Rad), which includes an initial genomic DNA digestion step. A primer pool targeting 22 lncRNAs and 2 reference genes was used (Supplementary Table 1). The reverse transcription and preamplification reactions were conducted in a thermal cycler following the manufacturer’s protocol, including a reverse transcription step and 14 preamplification cycles. Relative quantification of circulating lncRNAs was performed using real-time PCR (qPCR) with customized 384-well PrimePCR™ plates (Bio-Rad), containing pre-dispensed primers for 22 lncRNAs and 2 reference genes. Each reaction contained 1 μL of preamplified cDNA, 5 μL of SsoAdvanced™ Universal SYBR® Green Supermix (Bio-Rad), and 4 μL of nuclease-free water, for a total volume of 10 μL. Amplification was conducted on a CFX384 real-time thermal cycler (Bio-Rad) with the following cycling conditions: 95 °C for 2 min, followed by 40 cycles of 95 °C for 5 s and 60 °C for 15 s. A melting curve analysis was included to verify amplicon specificity. Relative expression levels were calculated using the 2^(−ΔCt) method, with normalization to the geometric mean of the reference genes, determined using the BestKeeper algorithm. Search and selection of lncRNAs The 22 lncRNAs were selected based on a literature search conducted in early 2022. The search strategy included the use of keywords and Boolean operators such as lncRNA, type 2 diabetes mellitus, non-diabetics, pre-diabetes, and subjects diagnosed with diabetes. Studies performed in both animal and human models were considered, and searches were carried out using the PubMed and Scopus databases. The inclusion criteria were: (i) articles published within the last 5 years (2018–2022); (ii) studies published in Q1-ranked journals (IF 2022, WOS); and (iii) evidence demonstrating an association of lncRNAs with the regulation of key biological processes, including inflammation, oxidative stress, insulin secretion, β-cell function, insulin resistance, and glucose metabolism. Preference was given to studies using plasma samples, although some relevant studies employing tissue samples were also included. After this preliminary selection, the LncExpDB version 2.0 database was used to confirm whether the selected lncRNAs targeted genes involved in processes such as inflammation, oxidative stress, insulin secretion, β-cell function, insulin resistance, and glucose metabolism. Statistical analysis Continuous variables are expressed as mean ± standard deviation (SD), while categorical variables are presented as absolute numbers and percentages. The normality of continuous variables was assessed using the Shapiro–Wilk test. Depending on the distribution, comparisons between two groups were performed using either the Student’s t-test (for normally distributed variables) or the Mann–Whitney U test (for non-normally distributed variables). Dimensionality reduction and discrimination among groups were analyzed using Principal Component Analysis (PCA) and Partial Least Squares-Discriminant Analysis (PLS-DA). A Random Forest model was used to identify the most important lncRNAs associated with disease development, with several adjustments, including class weight adjustment for Incident:No-incident (1:3), ntree = 500 and mtry = sqrt(number of lncRNAs) = sqrt(22) ≈ 5. The Variable Importance Plot (VIP) displays the top lncRNAs ranked by their Mean Decrease in Gini index. The selection of relevant lncRNAs was based on their Mean Decrease Accuracy (MDA) index greater than 5.6 (corresponding to approximately the top ~ 35% of the importance distribution, given the maximum MDA of 8.7) and on significant baseline differences between incident and non-incident T2DM groups (p < 0.05). The predictive performance of the models was evaluated using Receiver Operating Characteristic (ROC) curve analysis, comparing five logistic regression models for the prediction of T2DM development during 5 years of follow-up, using class weight adjustment. A risk score integrating both lncRNAs and clinical variables was constructed based on a multivariate logistic regression model predicting the 5-year incidence of T2DM. The model included baseline clinical parameters (age, sex, BMI, fasting triglycerides, HDL-cholesterol, HbA1c, and statin use) along with the expression levels of selected lncRNAs (LINC01116, XIST, CASC2, and LINC01370) and class weight adjustment. To compute the lncRNA score, the absolute values of each variable was used, multiplied by its corresponding standardized z-value derived from the regression coefficients (β divided by the standard error) of the logistic regression model. The score was calculated as a continuous variable for each participant and subsequently categorized into tertiles to define low-, medium-, and high-risk groups. Overall probability and time-to-event analyses were performed using Kaplan–Meier survival curves, and differences between groups were evaluated using the log-rank test. Hazard ratios (HR) and 95% confidence intervals (CI) were calculated using multivariate Cox proportional hazards regression models, adjusted for age, sex, BMI, triglycerides, HDL-cholesterol, smoking status, statin use, and dietary intervention. Correlation analyses were conducted using Pearson’s correlation coefficient for normally distributed variables or Spearman’s rank correlation for non-parametric data. All statistical analyses were performed using R software (version 3.5.1), and a two-sided p-value of < 0.05 was considered statistically significant [19]. Results Long non-coding RNA profile in patients with incident type 2 diabetes mellitus The baseline characteristics of the study population, stratified by T2DM incidence [Incident-T2DM (N = 107) vs Non-Incident (N = 355)] after a median follow-up of 60 months, were previously published (Supplementary Table 2). Regarding the lncRNAs included in the present study, Table 1 presents the baseline normalized expression levels of lncRNAs between the non-incident and incident-T2DM groups and unsupervised PCA and OPLS-DA models were showed in Supplementary Fig. 1B. Notably, all significant lncRNAs were upregulated in the incident-T2DM group compared to the non-incident group, including ARAP1-AS2 (p < 0.001), XIST (p < 0.001), CASC2 (p < 0.001), GATA6-AS1 (p < 0.001), CDKN2B-AS1 (p = 0.034), LINC01116 (p < 0.001), LINC01370 (p < 0.001), among others. No significant differences in the lncRNAs studied were observed between subjects randomized to the Mediterranean diet and those to the low-fat diet prior to any dietary intervention (Supplementary Table 3). Table 1Baseline Long non-coding RNAs included in this study stratified by type 2 diabetes mellitus incidenceLong non-coding RNAsNon-incidentIncident-T2DMP valueN = 355N = 107ARAP1-AS20.105 (0.007–1.001)0.931 (0.110–3.064)< 0.001***C4orf421.077 (0.377–2.054)1.229 (0.601–2.992)0.080CASC20.015 (0.003–0.095)0.115 (0.018–0.535)< 0.001***CDKN2B-AS10.167 (0.011–0.716)0.266 (0.031–1.179)0.034*GATA6-AS10.045 (0.005–0.472)0.199 (0.041–1.887)< 0.001***H1957.132 (25.483–118.733)67.882 (27.911–140.733)0.375HOTAIR5.931 (1.967–16.661)6.130 (2.541–20.814)0.186KCNQ1OT15.942 (2.521–13.149)6.160 (2.216–16.240)0.954LINC005990.253 (0.043–0.934)0.635 (0.182–2.098)< 0.001***LINC011160.042 (0.002–0.685)0.447 (0.035–3.037)< 0.001***LINC013700.022 (0.005–0.280)0.277 (0.018–1.367)< 0.001***lnc-A2M-30.025 (0.001–0.373)0.088 (0.006–0.782)0.003**lnc-BTG2-20.079 (0.026–0.303)0.139 (0.041–0.539)0.048*MAPKAPK5-AS12.109 (1.148–4.365)2.768 (1.085–5.079)0.174MIR600HG0.285 (0.013–1.579)0.631 (0.045–2.920)0.020*MIR7-3HG0.190 (0.078–0.541)0.340 (0.122–1.360)0.004**NEAT13.532 (1.942–5.680)4.702 (2.589–7.180)0.005**NKILA0.180 (0.038–0.988)0.160 (0.038–2.419)0.621OLMALINC0.367 (0.112–0.928)0.588 (0.244–1.698)< 0.001***PDX1-AS10.015 (0.002–0.288)0.112 (0.010–0.883)0.001**TUG12.536 (1.212–4.090)2.542 (1.755–4.059)0.296XIST0.077 (0.010–0.420)0.374 (0.030–0.967)< 0.001***Data are presented as median (Q1–Q3) or n (%), and values were normalized by multiplying by a factor of 103. Long non-coding RNA expression was normalized using the BestKeeper method, with B2M and ACTB as reference genes. Groups were divided according to the presence of type 2 diabetes mellitus incidence during a follow-up period of 5 years. A Shapiro-Wilks test was conducted to determine whether the data followed a normal distribution. Depending on the results, either the t-test (for normally distributed variables) or the Mann–Whitney test (for non-normally distributed variables) was applied. Statistical significance is indicated by asterisks (***p < 0.001, **p < 0.01, *p < 0.05) Baseline Long non-coding RNAs included in this study stratified by type 2 diabetes mellitus incidence Data are presented as median (Q1–Q3) or n (%), and values were normalized by multiplying by a factor of 103. Long non-coding RNA expression was normalized using the BestKeeper method, with B2M and ACTB as reference genes. Groups were divided according to the presence of type 2 diabetes mellitus incidence during a follow-up period of 5 years. A Shapiro-Wilks test was conducted to determine whether the data followed a normal distribution. Depending on the results, either the t-test (for normally distributed variables) or the Mann–Whitney test (for non-normally distributed variables) was applied. Statistical significance is indicated by asterisks (***p < 0.001, **p < 0.01, *p < 0.05) Use of lncRNAs as potential biomarkers for predicting T2DM To identify the most predictive lncRNAs for T2DM incidence, a Random Forest model (ntree = 500) was used to rank lncRNAs based on their importance, measured by the MDA index (Fig.1). The model achieved an OOB error rate of 26.4%, with a confusion matrix showing 12/91 Incident and 261/280 No-incident samples correctly classified. The top four lncRNAs—XIST, LINC01116, CASC2, and LINC01370 -, showed an MDA index greater than 5.6 (~ 35%) (Fig.1) and were significantly different between the incident-T2DM and non-incident groups (p < 0.05; Table 1). Variable importance plot showing the top lncRNAs ranked by their Mean Decrease in Gini index from the Random Forest model predicting disease incidence. Each bar represents the importance score of an individual lncRNA in classifying subjects into Incident-T2DM vs Non-incident groups. The color gradient reflects the magnitude of the importance score, with darker colors indicating higher importance. This plot highlights the lncRNAs that contribute most significantly to the predictive performance of the model (ntree = 500 and mtry = 5). Participants were classified according to MDA levels, and the top 35% of the distribution (≥ 5.6 units and the maximum (LINC01116) was 8.7 units; CASC2 = 5.6, HOTAIR = 5.76, XIST = 7.57, and LINC01370 = 7.58) was selected as the high-MDA Group. Abbreviations: lncRNA: long non-coding RNA; MDA: Mean decrease accuracy; T2DM: type 2 diabetes mellitus To further evaluate the predictive capacity of these four lncRNAs, ROC curve analyses were conducted using logistic regression models to predict T2DM incidence at 5 years (Fig. 2, Supplementary Table 4 ). Accordingly, the model, which included clinical variables (age, sex, BMI, diet, HDL, smoking status, triglycerides, statin use), HbA1c and the four selected lncRNAs, achieved the highest AUC of 0.730 (95% CI 0.671–0.788), with a sensitivity of 0.747, specificity of 0.611. Statistical comparisons using the one-sided DeLong test confirmed that this model significantly outperformed a model with HbA1c alone (p = 0.015), a model with FINDRISC model (p < 0.001), a model with clinical variables only (p < 0.001), and a model with clinical variables plus HbA1c, (p = 0.014). ROC curves comparing five logistic regression models for predicting incident diabetes at 5 years. Model 1 includes only baseline HbA1c levels, with an AUC of 0.658 (0.596–0.720). Model 2 includes only the FINDRISC score, with an AUC of 0.610 (0.550–0.671). Model 3 includes clinical covariates such as age, sex, BMI, triglycerides, statin use, diet, HDL, and smoking, with an AUC of 0.642 (0.580–0.703). Model 4 extends Model 2 by including baseline HbA1c, with an AUC of 0.696 (0.636–0.755). Model 5 adds four selected normalized lncRNAs (XIST, CASC2, LINC01116, LINC01370) to Model 4, with an AUC of 0.730 (0.671–0.788). The inclusion of lncRNAs in Model 5 improved classification performance, as evidenced by ROC analysis. Statistical comparison showed that Model 5 significantly outperformed Models (p < 0.05, one-sided DeLong test). Abbreviation: AUC: Area under curve; BMI: Body mass index; Hb1Ac: Glycated haemoglobin; HDL: High-density lipoprotein; lncRNA: long non-coding RNA; ROC: Receiver Operating Characteristic curves Construction of an epigenetic long non-coding RNA score for predicting type 2 diabetes mellitus development To construct an epigenetic lncRNA score, a regression model was employed as previously described. This epigenetic score was subsequently used in Cox regression models, adjusted for clinical variables to evaluate its association with T2DM incidence based on score tertiles. Among participants following a Mediterranean diet, those in the low-score group (T3) had a significantly lower risk of developing T2DM compared to the high-score group (T1) (HR 0.37, 95% CI 0.24–0.59, p < 0.001). No significant difference was observed between the high- and medium-score groups (T1 vs. T2; p = 0.835) (Fig. 3A). Moreover, significant differences were found across score tertiles in the overall population (HR 0.55, 95% CI 0.41–0.75, p < 0.001) and for diet interaction (p = 0.025), but no significant differences were observed among participants following a low-fat diet (Supplementary Figs. 2A and 2B, respectively). Cox regression analyses were generated based on tertiles of lncRNAs score for the assessment of the risk of developing T2DM. A participants adhering to a Mediterranean diet (N = 246). The model was adjusted for BMI, sex, age, HDL-C, triglycerides, smoking status and statin use. The continuous blue line represent low (T1) and light blue represents medium (T2) score groups, while the continuous orange line represents the high score group (T3). B Pearson correlation analysis of patients following a low-fat (N = 216) and a Mediterranean diet (N = 246), evaluating correlation between baseline DI and lncRNA score. Abbreviation: BMI: Body mass index; DI: Disposition index; HDL: High-density lipoprotein; lncRNA: long non-coding RNA. T1: Tertile 1; T2: Tertile 2; T2DM: type 2 diabetes mellitus; T3: Tertile 3 To explore the biological relevance of the lncRNA score in T2DM incidence, a Pearson correlation analysis was performed to assess the relationship between lncRNA score and the DI. Among participants following a Mediterranean diet, a significant negative correlation was observed between the lncRNA score and baseline DI (r = − 0.18, p = 0.009). In contrast, no significant correlation was found among those following a low-fat diet (p = 0.272) (Fig. 3B). Discussion In this study, novel findings were presented regarding the association between lncRNAs and the development of T2DM in patients with coronary heart disease. The results suggest that circulating lncRNAs levels could potentially serve as valuable biomarkers for predicting the T2DM risk. A lower epigenetic score—constructed from the top four lncRNAs combined with clinical variables—was associated with a reduced likelihood of developing T2DM, particularly among individuals following a Mediterranean diet. This association may be partly explained by the observed link between a lower baseline score and improved insulin secretion capacity, suggesting a mechanism through which dietary intervention could enhance molecular mechanisms associated with insulin sensitivity. Overall, these findings support the potential utility of the lncRNA score not only in identifying individuals at higher risk of developing T2DM, but also in guiding personalized dietary strategies to reduce T2DM incidence in patients with coronary heart disease. T2DM is one of the leading causes of morbidity and mortality worldwide. A major challenge lies in the early identification of individuals at risk, as T2DM is often diagnosed only once clinical symptoms have already appeared. Therefore, identifying individuals with cardiovascular disease who are at high risk of developing T2DM is of critical clinical importance. In this context, multiple studies have demonstrated that lncRNAs are emerging as promising biomarkers for T2DM [21]. LncRNAs are typically defined as non-coding RNA transcripts longer than 200 nucleotides. Unlike protein-coding sequences, most lncRNAs display high cell-type specificity, participating in various physiological processes [22]. These characteristics makes lncRNAs highly attractive in predicting and monitoring the development of major metabolic diseases, including T2DM [23]. Consistent with these findings, the present study identified four lncRNAs (XIST, LINC01116, CASC2, and LINC01370) capable of differentiating between individuals who developed T2DM and those who did not, suggesting their potential use in predicting disease onset in patients with coronary heart disease. However, future work in patients with diagnosed T2DM could clarify whether these lncRNAs are linked to glycemic control, β-cell decline, or the development of micro- and macrovascular complications. Accordingly, several studies have reported that XIST is upregulated in T2DM, as demonstrated by the findings from this study. For instance, one study found increased XIST expression in PBMCs from patients with T2DM compared to healthy controls [24]. Another study observed higher XIST levels in patients with T2DM compared to non-diabetic patients with coronary artery disease [25], consistent with the findings from this study. In contrast, other studies reported decreased XIST expression in PBMCs from individuals with metabolic syndrome [26] as well as in patients with T2DM and type 2 diabetic peripheral neuropathy [27], suggesting a context-dependent or diseases stage-specific role of XIST in metabolic disorders. Mechanistically, XIST appears to contribute to T2DM-related complications, including diabetic cardiomyopathy [28] and diabetic nephropathy, by regulating specific miRNAs. In diabetic nephropathy, silencing XIST has been shown to protect against renal interstitial fibrosis by upregulating miR-93-5p expression [29], reduce cell proliferation and induce apoptosis via the miR-423-5p/HMGA2 pathway in high glucose-treated human kidney cells [30], as well as protect against renal injury through the upregulation of miR-15b-5p [31]. These findings suggest that targeting XIST could represent a promising therapeutic strategy for T2DM-related renal complications. As for the other LncRNAs identified in this study, CASC2 appears to be dysregulated in patients with T2DM with related-complications. Accordingly, one study found that CASC2 expression in serum and renal tissue was specifically downregulated in patients with T2DM and chronic renal failure, but not in those with other complications [32], consistent with findings from other studies [33, 34]. Mechanistically, CASC2 overexpression has been showed to reduce JNK1 phosphorylation, thereby inhibiting the JNK pathway [33]. Additionally, CASC2 may attenuate inflammation and fibrosis in diabetic nephropathy models, by modulating several metabolic signaling pathways [35–37]. In this study, although CASC2 expression was higher in the incident-T2DM group, further studies are required to elucidate its role in T2DM pathogenesis and assess its utility as a predictive biomarker. Finally, studies investigating LINC01370 and LINC01116 in the context of T2DM are limited. However, recent research has reported their involvement in cell proliferation [38, 39]. In particular, LINC01370 overexpression has been shown to reduce PI3K and p-AKT protein levels, which may be relevant in T2DM pathophysiology [39]. Findings from the present study suggest that these lncRNAs are dysregulated during T2DM development and may display predictive value. Nonetheless, further research is needed to validate these results and clarify their mechanistic and biological relevance. In any case, the score constructed from these four lncRNAs in this study demonstrated good predictive capacity to distinguish between the incident-T2DM and the non-incident groups. Current predictive models, such as the ADA risk score [5] and FINDRISC [4], may oversimplify the complexity of T2DM and fail to capture key molecular mechanisms. The findings from this study suggest that molecular biomarkers may provide complementary value to established clinical risk scores. For instance, a recent systematic review and meta-analysis including 17 studies and 4685 participants reported that lncRNAs had an AUC of 0.65 for the diagnosis of prediabetes [6]. In comparison, the lncRNA-based score developed in our study achieved an AUC of 0.73 for developing T2DM, underscoring the potential of lncRNAs as predictive biomarkers. In fact, in our study, the model based on lncRNAs + clinical variables was significantly (approx. 17%) better than the FINDRISC score. These results strongly support the need to formally evaluate whether incorporating lncRNA signatures into existing clinical risk algorithms could improve the prediction not only of incident T2DM but also of diabetes-related complications. Future studies integrating molecular, clinical, and lifestyle-related variables (such as diet and physical activity) are warranted to establish the clinical utility and generalizability of such combined models. It is worth mentioning that the findings from this study were obtained in a secondary prevention cohort, which may limit direct extrapolation. Future studies in primary prevention and in more diverse cohorts, including different ethnicities and geographical regions, will be essential to determine the broader applicability and clinical relevance of our results. Nevertheless, the biological pathways linking lncRNAs to metabolic dysregulation are likely to be relevant across different high-risk groups. The findings from the present study introduce an interaction between lncRNA and diet and lifestyle. Indeed, results from this study suggest that lncRNAs could contribute to T2DM risk assessment, particularly among individuals following a Mediterranean diet. Notably, strong evidence supports dietary intervention as an effective strategy to T2DM prevention. A review of three large U.S. cohorts concluded that adherence to a healthy diet, combined with other lifestyle factors such as regular physical activity, could prevent the majority of T2DM cases [40]. In this context, the Mediterranean diet emerges as a particularly attractive preventive approach. Beyond general dietary recommendations, predictive tools that help identify individuals who would benefit most from a Mediterranean diet Intervention are of particular clinical interest. In this study, the lncRNA score appears to be a useful tool for selecting individuals with higher likelihood of benefiting from this dietary therapy. Accordingly, a lower baseline lncRNA score was associated with a reduced risk of developing T2DM during the follow-up period, particularly among individuals following a Mediterranean diet. This score also correlated with improved insulin secretion, suggesting its potential utility in guiding personalized dietary strategies to reduce T2DM risk. Although these findings are promising, further validation in independent cohort is still required. Several limitations of this study should be acknowledged. First, the research was conducted within the context of a long-term, well-controlled dietary intervention in patients with established cardiovascular disease, which may not fully reflect real-life conditions. Second, the primary aim of the CORDIOPREV trial was not to assess the incidence of T2DM, which may limit the specificity and generalizability of the findings from this study. Although the analyses from this study were adjusted for potential confounders, the results should be interpreted with caution. Another limitation is that this study did not examine the role of lncRNAs in participants with diagnosed T2DM, which precludes conclusions about their prognostic value in this group. Given the relatively limited sample size, the risk of false positives was minimized by adjusting for confounding variables and supporting the conclusions through AUC metrics, random forest analysis and adjusted Cox analysis. Finally, further molecular and mechanistic studies are needed to better understand the role of lncRNAs in T2DM development, which could help clarify and reinforce the biological relevance. Overall, an epigenetic score based on circulating lncRNAs appears to be a promising diagnostic tool for T2DM risk. The combination of this score with a Mediterranean dietary intervention may modulate this risk, supporting the use of personalized dietary strategies for T2DM prevention in high-risk populations. Conclusion This study highlights the potential of circulating lncRNAs as epigenetic biomarkers for predicting the development of T2DM in patients with coronary heart disease. The integration of lncRNAs with clinical variables may enhance predictive accuracy. Furthermore, an epigenetic score based on lncRNA expression could serve as a valuable clinical tool to guide the selection of the most efficient dietary intervention, such as the Mediterranean diet, for secondary prevention of T2DM. Supplementary Information Supplementary Material 1. Supplementary Material 2. Supplementary Material 3. Supplementary Material 4. Supplementary Material 5. Author contributions HB and MJPV contributed equally to this work. JLM and OARZ equally contributing senior authors. HB, MJPV, JLM and OARZ wrote the draft manuscript. JLRC, ALM, JDTP, JFAD, APAL, PPM, JDL collected data and performed the classification of participants. OARZ performed the experiments. JLRC, ALM, JDTP, JFAD, APAL, PPM, JDL performed the medical revisions of participants and clinical databases, and HB, OARZ performed the statistical analysis. HB, MJPV, MMG, PPM, JLM and OARZ interpreted the data and contributed to the discussion. HB, MJPV, MMG, JLM and OARZ contributed to the writing of the manuscript and revised it critically for important intellectual content. JLM and OARZ 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. Funding We would like to express our gratitude to all the subjects who participated in the study, and the Córdoba branch of the Biobank of the Sistema Sanitario Público de Andalucía (Andalusia, Spain) for providing the human biological samples. We would also like to thank the EASP (Escuela Andaluza de Salud Publica), Granada, Spain, which performed the randomization process for this study. The CORDIOPREV study is supported by the Fundación Patrimonio Comunal Olivarero, Junta de Andalucía (Consejería de Salud, Consejería de Agricultura y Pesca, Consejería de Innovación, Ciencia y Empresa), Diputaciones de Jaén y Córdoba, Centro de Excelencia en Investigación sobre Aceite de Oliva y Salud and Ministerio de Medio Ambiente, Medio Rural y Marino, Gobierno de España; Ministerio de Economía y Competitividad (AGL2012/39615, PIE14/00005, and PIE 14/00031 to JL-M; AGL2015-67896-P to JL-M); Consejería de Innovación, Ciencia y Empresa, Proyectos de Investigación de Excelencia, Junta de Andalucía (CVI-7450 to JL-M); and by the Fondo Europeo de Desarrollo Regional (FEDER), JPI HDHL-NutriCog (PCIN-2016–084 to JL-M). The present study is funded by ISCIII through the Acción Estratégica en Salud 2022, FIS Project PI22/01020 to OAR-Z. HB is supported by a “Sara Borrell” postdoctoral contract (CD22/00053) from the Instituto de Salud Carlos III—Madrid (Spain), “Financiado por la Unión Europea – NextGenerationEU” y mediante el Plan de Recuperación, Transformación y Resiliencia. H.B. also received funding from the project ‘PI24/00061’, funded by the ‘Instituto de Salud Carlos III (ISCIII)’ and co-funded by the European Union. This study was supported by “Centro de Investigacion Biomédica en Red Fisiopatología de la Obesidad y Nutricion”, which is an initiative of the “Instituto de Salud Carlos III” (ISCIII) of Spain, financed by the European Regional Development Fund, “A way to make Europe”/ “Investing in your future”. Oriol Alberto Rangel Zúñiga is supported by a Nicolas Monardes Contract (C1-0004–2024) funded by Consejería de Salud y Consumo, Junta de Andalucía. Data availability Collaborations with the CORDIOPREV Study are open to Biomedical Institutions, always after an accepted proposal for a scientific work. Depending on the nature of the collaboration, electronic data, hard copy data, or biological samples should be provided. All collaborations will be made after a collaboration agreement. Terms of the collaboration agreement will be specific for each collaboration, and the extent of the shared documentation (ie, deidentified participant data, data dictionary, biological samples, hard copy, or other specified data sets) will be also specifically set on the light of each work. Declarations Competing interests The authors declare no competing interests. Footnotes Publisher's Note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Hatim Boughanem and María José Párraga-Viúdez have contributed equally to this work and are first authors. José López-Miranda and Oriol Alberto Rangel-Zuñiga have contributed equally to this work and are joint senior authors. Contributor Information José López-Miranda, Email: md1lomij@uco.es. Oriol Alberto Rangel-Zuñiga, Email: oriol.rangel@imibic.org. References Associated Data Supplementary Materials Supplementary Material 1. Supplementary Material 2. Supplementary Material 3. Supplementary Material 4. Supplementary Material 5. Data Availability Statement Collaborations with the CORDIOPREV Study are open to Biomedical Institutions, always after an accepted proposal for a scientific work. Depending on the nature of the collaboration, electronic data, hard copy data, or biological samples should be provided. All collaborations will be made after a collaboration agreement. Terms of the collaboration agreement will be specific for each collaboration, and the extent of the shared documentation (ie, deidentified participant data, data dictionary, biological samples, hard copy, or other specified data sets) will be also specifically set on the light of each work.