Electronic supplementary material ( ESM )

DNA was extracted from 25 mg of snap frozen liver using the QIAmp mini kit (QIAGEN, Hilden, Germany) as indicated by the manufacturer with minor changes in volume of added PBS (80 μl) and buffer ALT (180 μl) for homogenization. After homogenization (Bead Ruptor Omni International) and addition of protease K, the homogenate was stored at room temperature for 11-14 days prior final extraction steps, which is in accordance with the instructions of the manufacturer (Qiagen) and DNA methylation was shown to be stable under such storage conditions [1, 2].

DNA was extracted from 25 mg of snap frozen liver using the QIAmp mini kit (QIAGEN, Hilden, Germany) as indicated by the manufacturer with minor changes in volume of added PBS (80 µl) and buffer ALT (180 µl) for homogenization. After homogenization (Bead Ruptor Omni International) and addition of protease K, the homogenate was stored at room temperature for 11-14 days prior final extraction steps, which is in accordance with the instructions of the manufacturer (Qiagen) and DNA methylation was shown to be stable under such storage conditions [1,2].
All primer sets, including an assay for the analysis of CpG1 to CpG5, a reverse assay for CpG6 and a third primer assay for cg25924746, were designed using the PyroMark Assay Design 2.0 software. Bisulphite PCR was performed with the PyroMark PCR kit with 15 to 20 ng of input bisDNA for 40 cycles. Bisulphite-treated control DNA with known methylation quantities (EpiTect control DNA, QIAGEN, Hilden, Germany) was used for assay establishment and validation. Furthermore, systematic quality controls were performed before pyrosequencing including tests for primer-dimers and primer self-annealing. These quality controls included sequencing of i) reverse and sequencing primer w/o PCR product, ii) reverse primer only, iii) sequencing primer only, and iv) PCR product w/o primers. The inter-assay variance is below 3% and the intra-assay variance is below 4% for all assays. DNA methylation was measured by bisulphite pyrosequencing using the PyroMark Q48 and PyroMark Q48 Advanced reagents (QIAGEN, Hilden, Germany) according to protocol.

RNA isolation and gene expression analysis
Total RNA was extracted from 25 mg of snap frozen liver using the Mirneasy mini kit (QIAGEN, Hilden, Germany) and quantified spectrometrically. 2 µg of RNA was reverse transcribed into cDNA using the SuperScript VILO cDNA synthesis kit (Invitrogen, Carlsbad, US).
For hepatic miRNA cDNA synthesis, 10 ng of total RNA was reverse transcribed with the TaqMan Advanced cDNA synthesis kit (Applied Biosystems, Foster City, US). Hsa-let-7e-5p expression (478579_mir, Applied Biosystems, Foster City, US) was measured by TaqMan Advanced miRNA assay and calculated using the ΔΔCt method with hsa-miR-24-3p (477992_mir, Applied Biosystems, Foster City, US) as housekeeping gene.

Measurement of serum miRNA
Serum miRNA was extracted from 200 µl frozen serum with the miRNeasy Serum/Plasma Advanced kit (QIAGEN, Hilden, Germany) using the recommended spike-in control cel-miR-39-3p (QIAGEN, Hilden, Germany). Serum miRNAs were reverse transcribed with the qScript microRNA cDNA synthesis kit (QuantaBio, Beverly, US) using 5 µl eluate (corresponding to 40 µl serum input) for cDNA synthesis in 20 µl cDNA reaction mixture. Gene expression was measured by qPCR using the FastStart Universal SYBR Green Master (Roche, Basel, CH), selfdesigned qPCR primer (ESM Table 2) and universal primer from the qScript kit. Each qPCR reaction contained 266 nl input serum and target expression was normalized to the spike-in control cel-miR-39-3p expression using the ΔΔCt method.

Methylation-sensitive luciferase reporter gene assay
Luciferase-plasmids with a CpG free backbone and CpG-free CMV promoter (pCpGL-CMV-Fluc [4]) where created with different inserts (ESMTable 2). Subsequent restriction digestion was performed with HindIII and NcoI (NEB, Ipswich, US) and ligation with double stranded oligonucleotides described above by Quick ligation protocol (M2200, NEB, Ipswich, US). Ligation mix was transformed into One Shot PIR1 chemically competent cells (Invitrogen, Carlsbad, US) with zeocin resistance. Plasmids were extracted by QIAprep Spin Miniprep kit (QIAGEN, Hilden, Germany). Consequently, plasmids were either in vitro methylated using SssI methylase (NEB, Ipswich, USA) or mock methylated (incubation without SssI). HepG2 ([HEPG2] (ATCC® HB8065™)) cells were purchased from ATCC (Manassas, US) and are regularly checked for mycoplasma in a standardised manner by qPCR test for mycoplasma performed under ISO17025 accreditation to ensure work with negatively tested cells. HepG2 cells were plated into 96-well plates one day prior transfection. At 60 -70% confluency cells were co-transfected as triplicates with 100 ng luciferase reporter plasmid and either 500 ng β-galactosidase control plasmid (pCMV-bGal) or 10 ng SV40 renilla control plasmid (pRL-SV40) per well using Lipofectamin 3000 (Invitrogen, Carlsbad, US). Cells were harvested after 24 h of incubation in 50 µl passive lysis buffer per well and luciferase activity was measured in 20 µl lysate by DualGlo kit (Promega, Madison, US). For β-galactosidase normalization, βgalactosidase assay was performed by β Galactosidase Enzyme Assay System (Promega, Madison, US). All luciferase assays were performed three times with three technical replicates each and the mean of all three experiments is shown.
6 Electrophoretic mobility shift assay 20 µl of binding reaction containing 10 mM Tris, 50 mM KCl, 1 mM DTT, 50 ng Poly dI dC, 10 µM ZnSO4, 0.2 ug BSA, 6.5 µg HepG2 nuclear extract (NE-PER Nuclear and Cytoplasmatic Extraction Reagents, Thermo Fisher) and 50 fmol biotinylated oligo (referred to as labelled DNA) were incubated for 30 min on ice prior gel loading on 6 % (wt/vol.) DNA retardation gel (Invitrogen, Carlsbad, US). For supershift assays, nuclear extract was incubated with a specific antibody (polyclonal IgG α SREBF1 PA1-46142, polyclonal IgG α Sp1 PA5-29165, both Invitrogen Antibodies, Carlsbad, US) for one hour on ice before gel-loading. Protein-DNA complexes were plotted on nylon membranes and detected via chemiluminescence using the Chemiluminescent Nucleic Acid Detection Module (Thermo Fisher, Waltham, US). Specificity of protein binding was tested for by competition experiments in which gradients of unbiotinylated oligos (referred to as unlabelled DNA) were added to the incubation mixture prior to gel loading. The competitive erasure of protein binding to IRS2 was repeated five times. The incubation with SREBF1-antibody and SP-1 antibody was repeated three times. The effect of SNP rs4547213 on protein binding was performed twice. One representative blot is shown.

Insulin treatment of HepG2 cells
HepG2 cells (200.000 cells per ml) were plated in 6-wells prior insulin treatment. At 50 -60 % confluency, cells were cultivated in high glucose (25 mmol/l) DMEM supplemented with 0.5 % (wt/vol.) BSA and insulin (100 nmol/l or 500 nmol/l). Cells cultivated in low glucose (8.3 mmol/l) DMEM supplemented with 0.5 % (wt/vol.) BSA served as control. After 24 h of treatment, cells were washed once with PBS and lysed in 700 µL QIAzol as indicated by Mirneasy mini kit (QIAGEN, Hilden, Germany). All subsequent miRNA-cDNA and mRNA-cDNA synthesis steps were performed as described above. Insulin treatment was performed three times in duplicates each and the mean of all experiments is shown.

General statistics
Prior statistical calculations, outliers were identified using the robust regression and outlier removal (ROUT) method of GraphPad Prism version 7 (GraphPad Software, La Jolla, US) applying Q = 1 %.
Relative gene expression was calculated by the ΔΔCt method, using a respective housekeeper gene for normalisation. Expression data was normalised to the respective control group. Results were either visualised as mean fold +/-SEM as group overview or normalised by minimum(min)/maximum (max) normalisation (1 -(xi -xmin)/(xmax -xmin)) to a scale from 0 (minimum expression) to 1 (maximum expression) for correlation analysis. Thereby xi is the current ΔCt value of a specific gene, xmin is the minimum ΔCt value of all measurements of a specific gene and xmax is the maximum ΔCt value of all measurements of a specific gene.
DNA methylation is shown as individual % methylation values which were calculated by the PyroMark Q48 Advanced Software (QIAGEN, Hilden, Germany). For plotting, each data point represents one analysed individual. The black line indicates the mean±SD.
Regression analyses were performed using MATLAB R2018a by fitting a linear regression model of the form y = β0 +β1X1 + … + βnXn with additional predictors β1 to βn (The MathWorks, Natick, US). Logistic regression analyses were similar performed using a multinomial logistic regression. Age, sex and BMI were used as predictors to control for the intracohort variance. Thereby generated pa<0.05 was assumed as significant.
Linear and rank correlation analyses were also performed by MATLAB R2018a using the corr function either as Pearson or as Spearman correlation. Two-sided Student's t test was used for the comparison of means for normally distributed measurements. Two-sided Wilcoxon rank sum test was used for nonparametric tests. One-way ANOVA was used to compare the means of more than two groups with a post hoc test to compare individual means. Post hoccalculated p values were adjusted for multiple testing by the FDR method of Benjamini and Hochberg and a q<0.05 was assumed as significant (GraphPad Prism, version 7; GraphPad Software, La Jolla, CA, USA). Moreover, we tested separately the influence of age and BMI by either correlating DNA methylation or gene expression with both factors. For the influence of sex on DNA methylation or gene expression, we performed a group stratification between male and female individuals and calculated a two-sided Wilcoxon rank sum test. A p<0.05 was assumed significant. P values were adjusted for multiple comparison by using the p.adjust function of R version 3.5.1 (RStudio, Boston, US) and controlling of the false discovery rate (FDR) with the method of Benjamini and Hochberg. Therefore, we applied a FDR = 15 % setting the level of significance to q<0.15.
After controlling for Hardy-Weinberg equilibrium, a χ 2 test with α = 0.05 was used to test for an association between the polymorphism and the incidence of type 2 diabetes. The effect of the polymorphism on IRS2 gene expression was assessed by calculation of a quantitative trait loci (eQTL) analysis performed by a linear mixed effect model (lmer) in R. The genotype of the polymorphism was used for the random effect on gene expression. Age, sex and BMI were used as cofactors (fixed effects).

ESM Tables
ESM Table 1: Clinical characteristics of subjects, stratified by medical examination into subjects with type 2 diabetes (T2D) or without type 2 diabetes (non-diabetic, ND). Significant differences in the cohort are indicated by bold p values (p<0.05).

ESM Table 5: p-values and q-values (FDR 5%) for the correlation analysis between hepatic IRS2 expression and downstream genes in the insulin signaling pathway.
All significant p-values (p<0.05) and q-values (q<0.05) are indicated in bold.  ). This is indicated by no difference in fold change expression relative to non-diabetic major allele carriers (CpG/CpG) after stratification for the genotype. Correlation analysis was performed on DNA methylation and dCt values first by Pearson correlation (p) and additionally by linear regression models to adjust for age, BMI and sex (pa). For visualization, gene expression was normalized by min/max normalization (1 -(xi -xmin)/(xmaxxmin)) to a scale from 0 (minimum expression) to 1 (maximum expression) and indicated as normalized expression on the axis. To assess the influence of the polymorphism and disease state, a two-way ANOVA was performed, *p<0.05. The influence of the polymorphism on disease incidence was calculated by logistic regression using age, BMI and sex as co-factors. Electrophoretic mobility shift assay shows a shift of biotinylated DNA oligonucleotides containing CpG1 to CpG6 after incubation with nuclear Extract of HepG2 cells (bIRS2+Protein, lane 2) which can be erased after incubation with different concentrations (12.5 pmol, 1.25 pmol, and 0.125 pmol, indicated as gradient) of unlabelled DNA (lanes 3-5 for unlabelled fullymethylated DNA and lanes 6-8 for unlabelled un-methylated DNA). Incubation of the binding reaction with antibody against SREBF1 (Ab α SREBF1, lane 9) or Sp1 (Ab α SP1, lane 10) for confirmation of proposed transcription factors results into intensity change of shifted biotinylated DNA. Control for nuclear extraction buffer only (Buffer, lane 11) shows no background signals. It is not possible to quantify the effects of competitive binding for fullymethylated and un-methylated DNA accurately due to signal saturation of the free DNA which serves as background control. The dashed lines indicate the cropped part which is missing in Fig. 3d.