Abstract Background Individuals with type 2 diabetes (T2D) face an increased mortality risk, not fully captured by canonical risk factors. Biological age estimation through DNA methylation (DNAm), i.e. the epigenetic clocks, is emerging as a possible tool to improve risk stratification for multiple outcomes. However, whether these tools predict mortality independently of canonical risk factors in subjects with T2D is unknown. Methods Among a cohort of 568 T2D patients followed for 16.8 years, we selected a subgroup of 50 subjects, 27 survived and 23 deceased at present, passing the quality check and balanced for all risk factors after propensity score matching. We analyzed DNAm from peripheral blood leukocytes using the Infinium Human MethylationEPIC BeadChip (Illumina) to evaluate biological aging through previously validated epigenetic clocks and assess the DNAm-estimated levels of selected inflammatory proteins and blood cell counts. We tested the associations of these estimates with mortality using two-stage residual-outcome regression analysis, creating a reference model on data from the group of survived patients. Results Deceased subjects had higher median epigenetic age expressed with DNAmPhenoAge algorithm (57.49 [54.72; 60.58] years. vs. 53.40 [49.73; 56.75] years; p = 0.012), and accelerated DunedinPoAm pace of aging (1.05 [1.02; 1.11] vs. 1.02 [0.98; 1.06]; p = 0.012). DNAm PhenoAge (HR 1.16, 95% CI 1.05–1.28; p = 0.004) and DunedinPoAm (HR 3.65, 95% CI 1.43–9.35; p = 0.007) showed an association with mortality independently of canonical risk factors. The epigenetic predictors of 3 chronic inflammation-related proteins, i.e. CXCL10, CXCL11 and enRAGE, C-reactive protein methylation risk score and DNAm-based estimates of exhausted CD8 + T cell counts were higher in deceased subjects when compared to survived. Conclusions These findings suggest that biological aging, as estimated through existing epigenetic tools, is associated with mortality risk in individuals with T2D, independently of common risk factors and that increased DNAm-surrogates of inflammatory protein levels characterize deceased T2D patients. Replication in larger cohorts is needed to assess the potential of this approach to refine mortality risk in T2D. Supplementary Information The online version contains supplementary material available at 10.1186/s12933-024-02351-7. Keywords: Type 2 diabetes, Epigenetic clocks, DNA methylation, PhenoAge, DunedinPoAm Background Type 2 diabetes (T2D) is an age-related metabolic disorder characterized by chronic hyperglycemia and insulin resistance, with a constantly increasing incidence and prevalence [[55]1]. T2D, similarly to others age-related diseases, is a multifactorial disease resulting from a combination of genetic and environmental factors and characterized by a pervasive status of low-grade inflammation which accelerates the onset of diabetic complications, i.e. cardiovascular diseases, nephropathy, retinopathy [[56]2, [57]3]. Individuals with T2D have an increased mortality risk compared with the general population, not fully captured by conventional risk factors, e.g. HbA1c, LDL cholesterol, and blood pressure [[58]4, [59]5]. Previous attempts of improving mortality risk stratification relied on the assumption that T2D might be considered a condition of accelerated aging, with multiple surrogate markers sustaining this hypothesis [[60]6–[61]9]. Among these biomarkers, DNA methylation (DNAm) emerged as a measure of biological age, leading to the development of the so-called “epigenetic clocks”, that are capable of providing a measurement of the quality of the individual aging process, prevalently expressed as age acceleration. Generally, negative values of age acceleration indicate healthy aging, while positive values reflect unhealthy aging and are commonly observed in age-related diseases [[62]10, [63]11]. Specifically, the first-generation “epigenetic clocks”, i.e. Hannum and Horvath clocks, were built using chronological age as output variable but had a poor performance in predicting morbidity and mortality outcomes [[64]12]. More recent approaches, using mortality as the key variable to build the clock and incorporating data from multiple sources, were instead associated with a range of age-related endpoints [[65]13, [66]14]. Indeed, age acceleration metrics are predictors of disease risk, morbidity and mortality in large cohorts of healthy subjects [[67]11, [68]15–[69]19]. In addition, other studies support the possible usefulness of DNAm-derived biological age in improving 10 year risk prediction for T2D [[70]20] or their ability to detect diabetes complications [[71]21, [72]22]. However, no study explored the ability of these tools to predict mortality specifically in subjects with T2D and independently of canonical risk factors. To explore the possibility that DNAm-derived biological age is associated with mortality independently of common risk factors in subjects with T2D, we leveraged a well-characterized cohort of individuals with T2D who were followed-up for 16.8 years to create two groups, one of deceased individuals and the other of survived patients, fully matched for common risk factors. We performed a genome wide DNAm analysis of peripheral blood leukocytes to explore differentially methylated genes related to death in T2D, test the ability of a chosen set of existing tools estimating biological aging from DNAm to predict mortality, and assess the differences between deceased and survived diabetic patients in DNAm-inferred levels of selected inflammatory proteins and in predicted blood cell counts with a known pathophysiological role in T2D. Methods Patient selection Patients were retrieved from a previously characterized cohort of 568 patients affected by T2D [[73]6]. Subjects were recruited between 2003 and 2006 in sites located within the Marche Region, Italy, according to the following inclusion criteria: clinical diagnosis of T2D established according to the American Diabetes Association guidelines [[74]23] from at least 3 years, age ranging from 55 to 70 years, HbA1c between 6.0 and 8.0%, BMI < 35 kg/m^2, eGFR > 45 mL/min, no current smoking, and no history of previous major adverse cardiovascular events (MACE), including non-fatal myocardial infarction or stroke. The presence of T2D complications, i.e. retinopathy, nephropathy, neuropathy, MACE, and atherosclerotic vascular disease was established as previously described [[75]6]. All participants were of European ancestry. Among the 181 subjects that met the inclusion criteria, 49 died during the 16 year follow-up period. A propensity score was calculated for each patient using a logistic regression model with baseline variables that potentially influenced the outcome, i.e. sex, age, HbA1c, eGFR, hs-CRP, LDL-cholesterol, disease duration, and BMI. We then matched 28 patients that were deceased during the follow-up period—14 males and 14 females—one-to-one with survived patients by propensity score matching using the nearest neighbor matching method of the R package MatchIt, version 4.5. The study was approved by the Institutional Review Board of IRCCS INRCA hospital (Approval No. 34/CdB/03). Written informed consent was obtained from each participant in accordance with the principles of the Declaration of Helsinki. DNA Extraction and methylation assay DNA extraction was performed using QIAamp DNA Blood Mini Kit (Qiagen) in spin procedure according to the manufacturer’s instructions. In brief, 200 µL of whole blood samples were lysed with 20 µL of Qiagen Protease in presence of Buffer AL incubating for 10 min in 56 °C. Samples were purified on mini spin columns with two consecutive washes with Buffers AW1 and AW2. DNA were unbound from membranes by 10 min incubation (room temperature) and elution in 200 µL of Buffer AE, and they were stored at 4 °C until quality control. Extraction yield was estimated using a Qubit 3.0 Fluorometer with dsDNA BR Assay Kit (Thermo Fisher Scientific) and samples were normalized to 1000 ng in 45 µL with ddH2O. Genomic DNA was bisulfite-converted using the EZ-96 Deep Well DNA Methylation Kit (Zymo Research) and analyzed using the Infinium Human MethylationEPIC v1.0 BeadChip (Illumina) according to the respective manufacturer’s instructions. All processing steps were performed with accurate randomization of the samples and phenotypic groups. Data preprocessing Raw idat files obtained from Illumina array run were preprocessed in Linux environment using bioinformatic pipeline implemented in R (version 3.6.3). This workflow included quality control, normalization, cleansing and filtering steps according to the recommendations of Maksimovic et al. [[76]24]. Briefly, for each sample we checked the quality by calculating its mean probe detection p-value and verifying that it reached the statistical significance (< 0.05). Data was normalized using noob background correction with dye-bias normalization (minfi R package, version 1.32.0) [[77]25]. We filtered out the probes which presented detection p-value > 0.01 in at least one of the samples, those located on sex chromosomes and those mapping to SNPs. Additionally, we excluded non-specific, cross-reactive, variant-containing, masked from mapping and multiple alignment probes according to the recently published recommendations regarding the Illumina arrays [[78]26–[79]29]. Only CpG sites that did not have bi- or tri-modal distribution in any of the sex groups of survived patients were considered. Eventually, for all successfully assessed probes we calculated beta values that express DNA methylation levels as a ratio of methylated to unmethylated alleles intensities (with 0 corresponding to totally unmethylated and 1—to totally methylated states) and used them in subsequent and differential and epigenetic estimates analysis. Differential methylation analysis CpG sites associated with diabetic patients’ condition (survived or deceased) were identified generating multiple linear models with robust regression fitting (limma R package, version 3.42.2) as previously described [[80]30]. The models were corrected for chronological age, sex, and presence of complications, DNA-methylation based estimates of blood cell counts (naive CD8 + T cells, CD4 + T cells, exhausted cytotoxic CD8 +, CD28−, and CD45R− T cells, natural killer cells, granulocytes, and plasma blasts) obtained with Horwath’s New DNA Methylation Age Calculator ([81]https://dnamage.genetics.ucla.edu/) and Illumina array batch. Differentially methylated positions (DMP) were identified selecting CpGs that i) reached significant p-value after Benjamini–Hochberg adjustment for multiple tests at statistical significance level of 0.05 and ii) presented absolute value of methylation difference between two compared groups above 5%. Differentially methylated regions (DMR) were detected using Comb-p approach [[82]31] which searches for associations using meta-analysis integrating nominal p-values of neighboring CpG sites that were previously estimated with linear models. Regions that reached adjusted combined p-value below 0.05 were considered as significantly associated with death in T2D. Differentially variable positions (DVPs) were revealed performing the methylation absolute deviation analysis as implemented in varFit() function of missMethyl R package, version 1.20.4. CpGs which demonstrated absolute value of variance ratio between two groups after logarithmic transformation above 2 and which presented nominal p-value below 0.001 were considered as DVPs. Pathway enrichment analysis We performed a pathway enrichment analysis to get an insight on functional significance of observed methylation changes. For this purpose all identified DMPs were mapped to genes and created list of emerged unique genes was uploaded to Enrichr web-based tool [[83]32, [84]33] and we used KEGG database [[85]34] for pathway annotation. In the analysis we focused on the pathways for with Fisher’s exact test p-values < 0.05. Epigenetic estimates analysis Whole-genome methylation data was used to evaluate a battery of DNAm estimates including i) predictors of biological aging, ii) biomarkers of plasma proteins, iii) signature of chronic low-grade inflammation based on C-Reactive protein (CRP) and iv) estimates of blood cell counts. Table [86]S1 provides a detailed list of DNAm variables that were assessed in this study, with short descriptions, indications of respective references and links to source scripts used for