In this study, we used the Leipzig Obesity BioBank (LOBB; https://www.helmholtz-munich.de/en/hi-mag/cohort/leipzig-obesity-bio-bank-lobb)comprising intra-individually paired SAT and OVAT samples organized in several sub cohorts (total N=962). All patients were extensively clinically phenotyped regarding obesity, metabolic variables and anthropometric traits as described by Blüher (2020) and Klöting et al. (2010) (21, 22). Adipose tissue samples were collected during elective laparoscopic abdominal surgeries as described previously (23). Exclusion criteria comprised individuals under the age of 18, chronic substance or alcohol misuse, smoking within the last year leading up to surgery, acute inflammatory diseases, end-stage malignant diseases, uncontrolled thyroid disorder, and Cushing’s disease. Three sub cohorts of LOBB were considered: a cross-sectional cohort (CSC, N=839), a two-step bariatric surgery cohort (BSC, N=52) and a cohort comprising individuals with metabolically healthy vs. unhealthy obesity (MHUO, N=71). The CSC cohort comprises 818 subjects with obesity (72% women, 40% T2D, mean ± sd age 46.9 ± 11.8 years; mean ± sd BMI 48.9 ± 8 kg/m² ) and 21 subjects without obesity (62% women, 17% T2D, mean ± sd age 49 ± 13.5 years; mean ± sd BMI 27 ± 2.5 kg/m²). The BSC cohort consists of 52 individuals that have undergone a two-step bariatric surgery at two time points and who were considered morbid obese at the first surgery. Only individuals who lost more than 10 kg of weight were considered. The first time point (first surgery) is labelled pre-surgery (70% women, 58% T2D, mean ± sd age 44.5 ± 9.6 years; mean ± sd BMI 54.3 ± 10.6 kg/m²), the second time point (second surgery) is labelled post-surgery (50% T2D, mean ± sd age 46.7 ± 9.6 years; mean ± sd BMI 40.2 ± 7.5 kg/m²). Individuals lost 41.1 ± 21.2 kg between the two surgeries. The MHUO cohort entails 30 obese individuals who are insulin sensitive (IS; 71% women, mean ± sd glucose infusion rate (GIR) clamp 89.5 ± 9.3 mmol/kg/min, mean ± sd age 38.5 ± 11.4 years; mean ± sd BMI 44.5 ± 7.1 kg/m², 10% T2D) and 41 individuals who are insulin resistant (IR; 73% women, mean ± sd GIR clamp 28.7 ± 12–6 mmol/kg/min, mean ± sd age 50.4 ± 8 years; mean ± sd BMI 47.2 ± 6.9 kg/m²) 61% T2D). The cohort participants were stratified into metabolically healthy and unhealthy obesity groups according to previously published parameters. A detailed description of the MHUO cohort is provided in Reinisch et al., 2025 (31).

Library preparation and RNA sequencing (RNA-seq) were carried out as previously described (24). Briefly, RNA was isolated from adipose tissue samples using the SMARTseq (Switch Mechanisms at the 5’ End of RNA Templates Sequencing) protocol (25). Sequencing libraries were generated and subjected to single-end sequencing on a NovaSeq 6000 platform at the Functional Genomics Centre Zurich.

Raw sequencing reads were processed using Fastp (v0.23.4) for adapter removal and quality trimming (26). A sliding window approach was applied, trimming the read ends when the average quality within a four-base window dropped below a Phred quality score of 20. Reads shorter than 18 bases after trimming were discarded. Trimmed reads were aligned to the human reference genome (GRCh38, GENCODE release 47) using STAR (v2.7.11b), allowing up to 50 multiple alignments per read (27). Gene-level quantification was performed using featureCounts (v1.5.3), with fractional assignment for reads mapping to multiple genomic locations (28). Count data were transformed using variance-stabilizing transformation (VST) via the DESeq2 R package v1.48.1, and expression values were adjusted for age, sex and proportion of reads assigned to features (exon mapping rate) using the r package limma v 3.64.0 before further analysis (29, 30).

Analysis was conducted in R 4.5.0. Statistical significance was assessed using Wilcoxon tests implemented via the rstatix R package (v0.7.2). For comparisons involving intra-individually paired samples (pre- and post-intervention measures from the same individual; tissue comparisons), the paired Wilcoxon signed-rank test was applied. For comparisons between independent groups, the unpaired Wilcoxon rank-sum test was used. P-values were adjusted for multiple testing using the Bonferroni correction method. Spearman’s rank correlation coefficients were calculated to assess monotonic associations between variables. Confidence intervals for the correlation coefficients were computed using the standard Fisher z-transformation. All correlation analyses were performed using the RVAideMemoire R package (v0.9-83-11).

To assess whether genes encoding for the m⁵C regulatory machinery are expressed in an adipose tissue depot-specific manner, we evaluated whether their expression levels differ significantly among our individuals with obesity of the cross-sectional cohort (N=818). Multiple writers (DNMT1, DNMT3A, DNMT3B, DNMT2, NSUN2, NSUN3, NSUN4, NSUN5, NSUN6, and NSUN7; all adj. p < 0.001; Figure 1A ), readers (MBD2, MBD3, MBD4, NEIL1, NTHL1, SMUG1, TDG, UHRF1, UNG, YBX1, and ZBTB33; all adj. p<0.001; Figure 1B ) and all of the known erasers (TET1, TET2, and TET3; all adj. p<0.01; Figure 1C ) are indeed significantly different expressed in SAT vs. OVAT ( Supplementary Table 1 ). Except NSUN5, the writers had a significantly lower expression in OVAT compared to SAT. This may hint at a potentially decreased deposition of the m⁵C mark on RNA molecules in OVAT. Moreover, in contrast, most of the genes encoding for reader proteins (MBD3, UNG, SMUG1, NEIL1, NTHL1, MBD4, and ZBZB33), which are generally important for RNA metabolism and function, were significantly upregulated in OVAT. The three erasers removing m⁵C from RNA molecules were downregulated in OVAT compared to SAT.

Furthermore, in OVAT, the DNMT3A writer gene (adj. p < 0.001) and the eraser enzymes TET3 and TET2 (adj. p<0.01) were significantly downregulated in individuals with obesity (N=818) compared to non-obese controls (N=21). Conversely, several reader genes (MBD2, SMUG1, and MBD4) were upregulated (adj. p < 0.05) in individuals with obesity ( Supplementary Figure 1 ; Supplementary Table 1 ). In SAT, the eraser enzymes TET1 and TET2 (adj. p<0.001) were downregulated in obese individuals, along with the reader UHRF2 and the writer NSUN3 (adj. p<0.01). The writer DNMT1 (adj. p<0.01) and the reader MBD3 (adj. p<0.05) were upregulated in obese individuals ( Supplementary Figure 2 ; Supplementary Table 1 ).

Taken together, these data from our large cross-sectional cohort provide sound evidence for adipose tissue depot-specific gene expression of m⁵C regulating genes. Some regulatory genes are differentially expressed between non-obese subjects and individuals with obesity, while certain patterns such as decreased eraser activity among obese individuals are observed across multiple depots.

We observed multiple correlations (p<0.05) between the expression levels of m⁵C methylation regulators and clinical variables related to obesity within the cross-sectional cohort ( Figure 2 , Supplementary Table 2 ). Most of these correlations with anthropometric variables were identified for body fat (N=292). In OVAT, NSUN2 (writer, ρ=-0.15, p=0.012) and MBD1 (reader, ρ=-0.16, p=0.007) exhibited weak negative correlations, while ALYREF (reader, ρ=0.12, p=0.041) showed a positive correlation with body fat ( Figure 2A ). In contrast, the SAT samples showed stronger negative correlations between body fat and several methylation regulators, including the writers NSUN2 (ρ=-0.14, p=0.016), NSUN4 (ρ=-0.12, p=0.048), and NSUN6 (ρ=-0.13, p=0.027), as well as the readers UHRF2 (ρ=-0.26, p<0.001), TDG (ρ=-0.14, p=0.014), and MBD1 (ρ=-0.13, p=0.028), along with the eraser TET2 (ρ=-0.15, p=0.009). Positive correlations with body fat were observed for the writer DNMT1 (ρ=0.2, p<0.001) and the readers SMUG1 (ρ=0.16, p=0.007), NEIL1 (ρ=0.17, p=0.005), and NTHL1 (ρ=0.2, p<0.001) ( Figure 2B ).

Additionally, positive correlations were found between maximum adipocyte size in OVAT (N=75) and the erasers TET2 (ρ=0.26, p=0.027) and TET3 (ρ=0.26, p=0.024) suggesting that higher activity of these enzymes may be associated with larger fat cells. Conversely, SMUG1 showed a negative correlation (ρ=-0.24, p=0.041), indicating that increased SMUG1 expression might be linked to smaller adipocytes ( Figure 2A ). In SAT, ZBTB33 was positively correlated (ρ=0.28, p=0.016), while SMUG1 remained negatively correlated (ρ=-0.24, p=0.037) with the maximum adipocyte size (N=75) ( Figure 2B ).

In both adipose tissues, associations between metabolic parameters and m⁵C methylation regulators were observed. For example, in OVAT, DNMT3A (ρ=-0.16, p=0.007) and NEIL1 (ρ=-0.15, p=0.009) showed negative correlations with HOMA-IR (N=288), suggesting that higher activity of these genes may be linked to improved insulin sensitivity. Several methylation regulators also demonstrated correlations with circulating FGF21 (fibroblast growth factor 21) levels (N= 360). MBD3 (ρ=0.11, p=0.033) and SMUG1 (ρ=0.15, p=0.006) were positively associated with FGF21 levels, while DNMT2 (ρ=-0.13, p=0.014), MECP2 (ρ=-0.12, p=0.024), and NSUN6 (ρ=-0.11, p=0.034) exhibited weak negative correlations. These findings imply a potential epigenetic influence on FGF21 regulation, which could impact metabolic signaling pathways. Furthermore, resistin levels (N=420) were inversely correlated with several methylation regulators namely UHRF2 (ρ=-0.11, p=0.027), UNG (ρ=-0.17, p<0.001), NSUN2 (ρ=-0.14, p=0.004), and NSUN4 (ρ=-0.15, p=0.002), as well as positively associated with ALYREF (ρ=0.1, p=0.036), DNMT1 (ρ=0.11, p=0.028), DNMT3B (ρ=0.1, p=0.049), and TET3 (ρ=0.11, p=0.026) suggesting these enzymes may have a role in inflammation or adipokine regulation ( Figure 2A ). In SAT, several methylation regulators were also associated with HOMA-IR, including DNMT2 (ρ=-0.16, p=0.008), NEIL1 (ρ=-0.15, p=0.014), NSUN4 (ρ=-0.13, p=0.028), NSUN7 (ρ=-0.16, p=0.008), and YBX1 (ρ=0.14, p=0.022; N=288). Additionally, ALYREF (ρ=0.12, p=0.012) and ZBTB4 (ρ=-0.13, p=0.004) showed associations with HbA1c (N=450). Positive correlations with adiponectin levels were observed exclusively in SAT and involved MECP2 (ρ=0.22, p=0.047), MBD2 (ρ=0.24, p=0.031), and UHRF2 (ρ=0.26, p=0.018) ( Figure 2B ).

In summary, these findings highlight the potential role of specific m⁵C regulators in fat accumulation, adipocyte size, and obesity-related traits in a depot-specific manner, providing insights into potential epigenetic contributions to obesity and metabolic health.

Next, we tested whether transcriptional levels of the m⁵C regulatory machinery are affected by weight loss in the two-step bariatric surgery cohort and whether this relates to any correlation with clinical parameters. We found significant (adj. p < 0.05) differences in gene expression of four m⁵C regulators between the pre- and post-surgery timepoints (DNMT1, NSUN2, and YBX1 in OVAT, Figure 3A ; TET3 in SAT, Figure 3B ). All these regulators exhibited significantly lower expression levels after weight loss (post-surgery) compared to baseline (pre-surgery) values. An overview of all comparisons is provided in Supplementary Table 3 . No other m⁵C regulators were found differentially expressed.

The correlation analysis between delta (pre – post) changes in m⁵C regulator expression and delta weight loss-related phenotypes in our bariatric cohort revealed several associations (p<0.05) ( Supplementary Table 4 , Figures 3C, D ). In OVAT, several correlations were observed between the expression changes of several m⁵C regulators and measures of fat loss or lipid-related phenotypes. Specifically, decreases in NSUN4 (ρ=-0.33, p=0.018, N=52) and TET3 (ρ=-0.44, p=0.001, N=52) delta expression values were associated with greater reductions in EBMIL (excess body mass index loss), indicating that lower NSUN4 and TET3 levels post-surgery may be linked to more effective weight loss. Decreases in NSUN6 (ρ=-0.55, p=0.025, N=17) and TET2 (ρ=-0.49, p=0.049, N=17) expressions were associated with reductions in body fat. In addition, strong inverse correlations were observed between TET2 (ρ=-0.69, p=0.011, N=13), TDG (ρ=-0.56, p<0.05, N=13), and YBX1 (ρ=-0.62, p=0.029, N=13) expression and LDL (low-density lipoprotein) cholesterol. Regarding triglycerides, decreases in DNMT3B (ρ=-0.55, p=0.016, N=19) and TET2 (ρ=-0.6, p=0.008, N=19) delta expressions were linked to lower delta triglyceride levels. Conversely, an increase in UHRF2 (ρ=0.48, p=0.037, N=19) expression was positively correlated with triglyceride levels ( Figure 3C ).

In SAT, changes in ALYREF expression were positively correlated with measures of fat loss, including BMI (ρ=0.33, p=0.016, N=52), body weight (ρ=0.35, p=0.011, N=52), and EBMIL (ρ=0.29, p=0.036, N=52). These results suggest that increased ALYREF expression in SAT may be associated with greater reductions in body mass and enhanced metabolic activity during weight loss. In contrast to OVAT, the expression levels of several m⁵C regulators in SAT were associated with FPG and HOMA-IR. Specifically, NSUN1 (ρ=-0.36, p=0.016, N=44) and NTHL1 (ρ=-0.33, p=0.029, N=44) expression showed a negative correlation with FPG, indicating that higher NSUN1 and NTHL1 levels may be linked to lower FPG. Conversely, NSUN6 (ρ=0.35, p=0.022, N=44) and MBD2 (ρ=0.3, p=0.048, N=44) expression levels show a positive correlation with FPG. Further, UHRF1 expression showed a strong negative correlation (ρ=-0.59, p=0.049, N=12) with HOMA-IR, implying that higher UHRF1 levels might be associated with improved insulin sensitivity in SAT ( Figure 3D ).

This data illustrates that the gene expression profile of the m⁵C regulatory machinery is altered in the two-step bariatric surgery group and that such changes seem to be related to metabolic parameters. This inspired us to next evaluate the expression profiles in a cohort comprising metabolically healthy vs. unhealthy individuals with obesity.

To profile the expression landscape of m⁵C regulatory genes across various cell types, single-nucleus RNA sequencing (snucSeq) data from IS and IR morbidly obese individuals were analyzed ( Figure 4 ) (31). These data were obtained from Reinisch et al., 2025 (31), and include a subset of participants from our metabolically healthy versus unhealthy obesity (MHUO) cohort. In brief, snucSeq was performed on pooled adipose tissue samples from metabolically healthy and unhealthy obese individuals, comprising 41 OVAT and 36 SAT samples. Despite an overall lower gene expression of m⁵C regulators in SAT compared to OVAT, several m⁵C regulator genes displayed markedly elevated expression levels and were broadly distributed across various immune cell types (e.g., macrophages, monocytes, and T cells) and stromal cell types (e.g., adipocytes and mesothelial cells). Specifically, MECP2, MBD2, TET3, TET2, and DNMT3A were strongly expressed in macrophages, a key immune cell type involved in adipose tissue inflammation and metabolic dysfunction. These genes also showed high expression in monocytes, suggesting their broader involvement in innate immune regulation. Among the regulators, MECP2 exhibited the strongest and most widespread expression across virtually all cell types, suggesting a potentially global role in transcriptional and epigenetic regulation within adipose tissue. MBD2 showed the highest expression in adipocytes in both tissues, as well as in meMesoCs (mesenchymal-like mesothelial cells) and cMesoCs (mesothelial cell subtypes) in OVAT, indicating a possible role in modulating both lipid storage and mesothelial cell identity or function. TET1 and ZBTB4 are also stronger expressed in adipocytes ( Figure 4 ).

Analyses of m⁵C regulator RNAseq expression values between IS and IR samples of the MHUO cohort further ( Figure 5 , Supplementary Tables 5 , 6 ) supported the functional relevance of selected m⁵C regulators identified by snucSeq. For example, MBD2 expression is significantly reduced in omental adipose tissue of IR individuals compared to IS individuals ( Figure 5A ), suggesting a potential loss of epigenetic regulation in key stromal cell types associated with insulin resistance. In IS individuals OVAT, DNMT3A expression is significantly negatively correlated with circulating leptin levels (ρ=-0.52, FDR=0.003), suggesting a potential anti-inflammatory or metabolically protective role. Additionally, TET3 expression in IS individuals in both tissues positively correlated with chemerin levels (OVAT: ρ=0.34, p=0.049; SAT: ρ=0.54, p=0.003) and positively with IL-6 levels in SAT (ρ=0.37, p=0.043) consistent with its stromal expression and involvement in adipokine regulation. Correlations observed for MECP2 and ZBTB4 in both IS and IR individuals, further support their roles in glycemic and inflammatory control, in line with their expression across adipocytes and immune cell types ( Figure 5B ).

The role of the epitranscriptomic mark m⁵C has not been previously well described in the aetiology of obesity. A handful of studies reported a potential role of m⁵C in adipogenesis. Liu et al. (2021) provided insights into the molecular function of mRNA m⁵C in relation to the export of CDKN1A mRNA from the nucleus in controlling adipogenesis (19). They report the direct involvement of the writer and methyltransferase NSUN2 in regulating the m⁵C levels and the reader protein ALYREF to export CDKN1A mRNA to the cytoplasm (19). Furthermore, ALYREF was demonstrated to export YBX2 and SMO mRNA into the cytoplasm of porcine cells in a m⁵C-dependent manner to inhibit adipogenesis and promote myogenesis respectively (20). Hence, the findings reported by Liu et al. (2021) illustrate that known m⁵C regulators are involved in controlling adipogenesis, suggesting a possible link with obesity (19, 20). Our results provide novel insights in the gene expression profiles of writers, readers and erasers of m⁵C methylation in human adipose tissue and about depot-specific expression patterns in SAT vs. OVAT along with its adipose residing cell types in individuals with obesity, after weight loss and in individuals with metabolic alterations. In general, all except for NSUN5 of the m⁵C writers included in our study showed consistently lower gene expression in OVAT compared to SAT in the large cross-sectional cohort. The higher gene expression for m⁵C writers in SAT, especially the NSUN family with NSUN2, NSUN3, NSUN4, NSUN6 and NSUN7 all providing strong significances (p <0.001), is very interesting as higher levels of NSUN2 have been shown to inhibit adipogenesis (19, 20). This may indicate that depot-specific alterations in adipogenesis governed by m⁵C may be related to metabolic and anthropometric variables. Moreover, the writer NSUN3 was reported to catalyze the formation of m⁵C in mitochondrial tRNAs. (13). Given that mitochondrial dysfunction is known to be a consequence of obesity (32, 33), depot-specific NSUN3 gene expression may indicate at mitochondrial m⁵C potentially involved in obesity in a depot-specific manner (13). Moreover, one can speculate that a potentially higher activity of the writing machinery in SAT compared to OVAT may potentially lead to lower m⁵C deposition in OVAT. However, since our study lacks functional validation, more studies are warranted to test this. No causative conclusions can be drawn so far and the data presented in this study should therefore be interpreted with caution.

Moreover, our data indicate that weight loss (as demonstrated in our two-step bariatric surgery cohort) seems to be related to less pronounced depot-specific differences between SAT and OVAT. Four m⁵C regulators YBX1, DNMT1, NSUN2 in OVAT and TET3 in SAT, are differentially expressed between pre- and post-surgery, and their expression levels were all downregulated following weight loss. Interestingly, when analyzing cell-type specific gene expression by using snucSeq data, we discovered several genes strongly expressed in immune cells. The elevated expression of MECP2, MBD2, TET3, TET2, and DNMT3A in macrophages, particularly in omental tissue, supports the hypothesis of epigenetic reprogramming contributing to the development of insulin resistance, as macrophage-driven inflammation is a hallmark of IR-associated adipose remodeling.

The m⁵C readers MBD3, UNG, SMUG1, NEIL1, NTHL1, MBD4 and ZBTB33 show higher expression in OVAT compared to SAT. TDG, YBX1, MBD2 and UHRF1 showed higher expression levels in SAT compared to OVAT. NTHL1, primarily known for its role in DNA repair, has also been identified alongside eight other m⁵C readers as part of a specific m⁵C modification signature used to detect early acute myocardial infarction (34, 35). Similarly, TDG has been proposed as a diagnostic m⁵C-associated marker in idiopathic pulmonary fibrosis (36). Additionally, NEIL1, a key component of the base excision repair pathway essential for genomic integrity, is also implicated in various human diseases. Defects in NEIL1 have been associated with increased risks of cancer, diabetes and obesity (37). In the weight-loss two-step bariatric surgery cohort, the reader YBX1, DNMT1, writer NSUN2 and the eraser TET3 were the only regulators showing a significant differential gene expression between pre-and post-surgery timepoints and were observed downregulated following surgery. This data may hint at a potential impact of weight loss on the expression of these genes and the downregulation of YBX1 could result in decreased levels of its target proteins as it is known to actively stabilize mRNA through directly binding to m⁵C (38). However, no causative conclusions are possible due to the lack of functional validation in our study. Importantly, given that our cross-sectional cohort comprised only a limited number of non-obese individuals, our comparisons between obese vs non-obese individuals showing in general lower expression levels for the m⁵C machinery in subjects with obesity cannot be overestimated, and more data and larger cohorts are warranted in the future to draw any definite conclusion.

Furthermore, in the MHUO cohort MBD2 had a lower expression in IR individuals in OVAT compared to the IS counterparts. Given that MBD2 was predominantly expressed in adipocytes and mesothelial cells in the snucSeq data, this downregulation may reflect a loss of epigenetic repression mechanisms in key stromal compartments. Such a reduction could lead to increased transcription of pro-inflammatory or fibrotic genes, contributing to the dysfunctional remodeling of visceral fat observed in insulin resistance. These findings suggest a potential protective role of MBD2 in maintaining adipose tissue homeostasis, particularly in visceral depots.

Moreover, we observed higher expression levels in SAT compared to OVAT for the erasers TET1, TET2, and TET3, while the expression is lower across tissue depots among subjects with obesity. This may potentially hint at a general reduction of the eraser activity suggesting the possibility that individuals with normal weight have lower m⁵C levels compared to obese patients. The TET family is known to be involved in demethylation of both DNA and RNA m⁵C modification, in addition to their ability to yield 5-hydroxymethylcytosine (5hmC), 5-formylcytosine (5fC) and 5-carboxycytosine (5caC) through oxidation of m⁵C in DNA (39, 40). TET2 has previously been identified as an antiadipogenic demethylase in adipocyte differentiation in 3T3-L1 cells (41). Additionally, TET3 expression was decreased in SAT following weight loss. These findings are supported by Jung et al. (2023), who showed that TET3 is important for white adipose tissue development and expansion. Our results from the CSC cohort are in line with this, as we observed a higher TET3 expression in SAT (42). Jung et al. (2023), further proposed that TET3 regulates adipogenesis and diet-induced tissue expansion by modulating DNA methylation at important target genes in adipose precursor cells (42). Lastly, due to the involvement of the TET family in multiple forms of modification in DNA and RNA the results should be carefully interpreted as future validation studies are needed to confirm m⁵C methylation levels in the two depots investigated (39, 40).

Collectively, these results demonstrate a distinct depot- and cell-type-specific pattern of m⁵C regulator expression, with omental macrophages showing particularly elevated expression levels. This suggests a potential epigenetic axis through which visceral adipose tissue may contribute to insulin resistance in morbid obesity.

We find expression of multiple m⁵C regulators correlated with several clinical traits related to obesity and metabolism. In the CSC cohort, anthropometric traits like body fat were significantly correlated with a multitude of genes in both OVAT and SAT. This correlation was more widespread across several regulators in SAT compared to OVAT. Interestingly, NSUN2 was found to be negatively correlated with body fat for both depots. In line with this, we observed lower NSUN2 gene expression post-surgery and among subjects with obesity across both adipose tissue depots. These results align with the findings from Liu et al. (2021) showing that overexpression of NSUN2 corresponds to lower adipogenesis (19). In SAT, two additional NSUN family members (NSUN4 and NSUN6) are negatively correlated to body fat. NSUN4 have been linked to recruiting the degrading machinery for dsRNA in mitochondria to inhibit activation of the immune response due to faulty mitochondrial action (43). This may suggest a potential link between NSUN4 expression and the increase of immune cells in obesity. Furthermore, the three reader proteins NEIL1, NTHL1, SMUG1 and the writer DNMT1 all had a strong positive correlation in SAT with percentage body fat. The reader proteins are mainly known to be involved in the DNA base excision repair pathway, an important mechanism to be functional in an active depot (34). Moreover, in addition we observed strong correlations of multiple m⁵C regulators with mean and maximum adipocyte size suggesting a potential functional mechanism linking the m⁵C machinery to adipose biology. Furthermore, gene expression of multiple m⁵C regulators is correlated with important metabolic variables such as fasting plasma glucose and HOMA-IR across depots. However, only the SAT depot showed positive correlations for the three m⁵C readers MBD2, MECP2 and UHRF2 with adiponectin expression. Adiponectin is an adipokine with insulin-sensitizing properties and mainly originates from the SAT depot (44).

Although this is to our knowledge the first study exploring the relationship of m⁵C regulator gene expression and clinical variables related to obesity, our study has several limitations. First, two of the cohorts (BSC and MHUO) have relatively small sample size, which may limit the statistical power and generalizability of the findings described here. Second, the number of subjects with normal weight is relatively low in our cross-sectional cohort and our observed findings need to be interpreted with caution. Third, some of the genes identified by Song et al. (2022) to be members of the m⁵C regulating machinery, are also described as DNA m⁵C regulators (such as DNMT family and MBD family). Therefore, the observed correlation of gene expression with clinical traits may concern both DNA methylation and RNA methylation and should be carefully interpreted. Moreover, other genes may be involved in the regulation of RNA m⁵C, that were not included in the gene set identified by Song et al. (2022) and are therefore not analyzed in our study. Lastly, more functional data are therefore warranted to confirm or reject our results.