Dementia has become a major public health challenge. The number of dementia patients worldwide has reached 55 million. Dementia is a chronic neurodegenerative disease that is associated with various factors such as lifestyle (Gao et al., Reference Gao, Zhang, Song, Gan, Lin, Hu and Wu2024; Ye et al., Reference Ye, Sun, Wang, Khoo, Lim, Lu, Yu, Li, Maier and Feng2023), diseases (B. Zhang et al., Reference Zhang, Langa, Weuve, D’Souza, Szpiro, Faul, Mendes de Leon, Kaufman, Lisabeth, Hirth and Adar2023; Y. Zhang et al., Reference Zhang, Chen, Deng, You, He, Wu, Wu, Yang, Zhang, Kuo, Feng, Cheng, Suckling, David Smith and Yu2023), genetics (Loy et al., Reference Loy, Schofield, Turner and Kwok2014; Sassi et al., Reference Sassi, Guerreiro, Gibbs, Ding, Lupton, Troakes, Al-Sarraj, Niblock, Gallo, Adnan, Killick, Brown, Medway, Lord, Turton, Bras, Morgan, Powell, Singleton and ¼ Hardy2014), and obesity (Singh-Manoux et al., Reference Singh-Manoux, Dugravot, Shipley, Brunner, Elbaz, Sabia and Kivimaki2018; Stingl et al., Reference Stingl, Kullmann, Ketterer, Heni, Häring, Fritsche and Preissl2012). Lean body mass (LBM), as an indicator of obesity, is primarily composed of bone and muscle. LBM constitutes a large part of the human body and plays an essential role in various physiological processes (Wolfe, Reference Wolfe2006). Studies have found that LBM is associated with a reduced risk of mortality (Lee et al., Reference Lee, Keum, Hu, Orav, Rimm, Willett and Giovannucci2018), diabetes (Kalyani et al., Reference Kalyani, Tra, Egan, Ferrucci and Brancati2014), and various other diseases (S. Li et al., Reference Li, Jiao, Yang, Li, Zhang, Liu and Xue2023). Additionally, LBM has been shown to be associated with a reduced risk of cognitive dysfunction (Gong et al., Reference Gong, Tang, Chai, Qiao, Xu, Patel, Zhang and Zhou2023). Aging is often accompanied by a reduction in LBM and an increase in fat, leading to conditions like sarcopenia and sarcopenic obesity, which are linked to an elevated risk of cognitive impairment (Booranasuksakul et al., Reference Booranasuksakul, Macdonald, Stephan and Siervo2024; Hu et al., Reference Hu, Peng, Ren, Wang and Wang2022). However, the mechanism of the association between LBM and cognition remains unclear.
Previous research demonstrated that LBM was positively associated with cognition (Gong et al., Reference Gong, Tang, Chai, Qiao, Xu, Patel, Zhang and Zhou2023). LBM can affect the DNA methylation (DNAm) levels of genes. One genomewide DNA methylation analysis in Chinese monozygotic twins identified a causal effect of LBM on the DNAm of HELZ2 and CBLN1 genes (Tian et al., Reference Tian, Qiao, Han, Kong, Zhu, Xing, Duan, Li, Wang, Zhang and Wu2023). Moreover, DNAm can affect cognition. The DNAm levels of genes such as BDNF and CAPH2 can affect cognition. The methylation of the BDNF gene has been found to affect cognition, as well as change in methylation of CAPH2 that may predict the progression from amnestic MCI to Alzheimer’s disease (Ferrer et al., Reference Ferrer, Labad, Salvat-Pujol, Barrachina, Costas, Urretavizcaya, de Arriba-Arnau, Crespo, Soriano-Mas, Carracedo, Menchón and Soria2019; Hao et al., Reference Hao, Li, Han, Han and Cai2023). Based on these findings, we hypothesized that DNAm variation might mediate the association between LBM and cognition.
As shown above, DNAm can affect cognition. Additionally, DNAm can also affect LBM. One genomewide DNA methylation analysis also found that the DNAm of CPEB3, HELZ2 and CBLN1 genes had a causal effect on LBM (Tian et al., Reference Tian, Qiao, Han, Kong, Zhu, Xing, Duan, Li, Wang, Zhang and Wu2023). Furthermore, studies have found that LBM was associated with cognition. Based on these findings, we hypothesized that LBM may act as a mediator in the association between DNAm and cognition. However, current research comprehensively investigating the relationship between LBM, DNAm, and cognition remains limited.
Monozygotic twin pairs share the same genetic sequence and intrauterine environment, and similar growth environments. Consequently, using data from monozygotic twins may better control the effects of genetic variation and other potential confounding factors, thereby enhancing statistical power and reducing the required sample size (W. Li et al., Reference Li, Christiansen, Hjelmborg, Baumbach and Tan2018; Semenova et al., Reference Semenova, Vlasov, Partevian, Rosinskaya, Rybolovlev, Slominsky, Shadrina and Alieva2022). In this study, based on a sample of monozygotic twins from China, we performed mediation analysis to explore the mediating role of DNAm in the association between LBM and cognition, as well as the possible mediating role of LBM in the association between DNAm and cognition. The results will provide a theoretical basis for understanding the relationship between LBM, DNAm, and cognition, and for exploring the potential mechanisms underlying cognitive decline.
Methods
Participants
The twin samples were derived from the Qingdao Twin Registry, which was established in 1998. The participants who were pregnant or breastfeeding, unable to complete the assessment, or diagnosed with dementia were excluded. A total of 59 pairs of monozygotic twins with discordant cognition were finally included in this study.
This study conformed to the declaration of Helsinki, and all participants provided written informed consent. This study was approved by the Regional Ethics Committee of the Qingdao Center for Disease Control (Approval No.: 2012–01; Date: 2012-01-20).
Study Power Estimation
A previous study by W. Li et al. (Reference Li, Christiansen, Hjelmborg, Baumbach and Tan2018) has shown that in trait/disease-discordant twin design, when a trait/disease had a moderate and high heritability and there was a low correlation between environmental factors and DNAm, 80% statistical power would be achieved when the trait/disease-inconsistent twin sample size ranged from 22 to 63 (W. Li et al., Reference Li, Christiansen, Hjelmborg, Baumbach and Tan2018), which allowed for large sample size reductions compared to the ordinary case-control design. Existing research reported that heritability estimates of LBM in adults was 0.52 (Arden & Spector, Reference Arden and Spector1997) and 0.44 for cognition (Xu et al., Reference Xu, Sun, Duan, Ji, Tian, Zhai, Wang, Pang, Zhang, Zhao, Li, Gue, Hjelmborg, Christensen and Tan2015). For CpG sites exhibiting mediating effects, the median (interquartile range) of within-pair methylation correlations were 0.494 (0.476, 0.544) and 0.332 (0.197, 0.374), and the median (interquartile range) of environmental correlations with DNAm were 0.155 (0.148, 0.180) and 0.163 (0.093, 0.233). Therefore, according to the study by W. Li et al., our study based on nearly 60 pairs of monozygotic twins would achieve an 80% statistical power.
Cognition Aassessment
Cognition was assessed using the Montreal Cognitive Assessment (MoCA, Chinese version), which evaluates seven cognitive domains: visuospatial/executive function, naming, memory, attention, language, abstraction, and orientation. The total score is 30 points, with higher scores indicating better cognition (K.-L. Chen et al., Reference Chen, Xu, Chu, Ding, Liang, Nasreddine, Dong, Hong, Zhao and Guo2016). To control the impact of education, cognition scores were adjusted based on the level of education.
LBM Assessment
Trained investigators measured the participants’ height, weight, waist circumference, and other indicators. Each measurement was taken three times, and the average was used for subsequent analysis. This study used a body measurement prediction equation to calculate LBM. The equation was developed and validated using data from a large sample population, including gender, age, height, weight, and waist circumference. Research has shown that this equation has high predictive ability for LBM (Lee et al., Reference Lee, Keum, Hu, Orav, Rimm, Sun, Willett and Giovannucci2017). The LBM (kg) calculation formula was as follows:
$${\rm{Male}}:{\rm{ }}19.363{\rm{ }} + {\rm{ }}0.001{\rm{ \,}}^{*}{\rm{ age }}\left( {{\rm{years}}} \right){\rm{ }} + {\rm{ }}0.064{\rm{\, }}^{*}{\rm{ height }}\left( {{\rm{cm}}} \right){\rm{ }}$$$ + {\rm{ }}0.756{\rm{ \,}}^{*}{\rm{ weight }}\left( {{\rm{kg}}} \right){\rm{ }} - 0.366{\rm{\, }}^{*}{\rm{ waist circumference }}\left( {{\rm{cm}}} \right){\rm{ }}$$$ - {\rm{ }}1.007$$
$${\rm{Female}}:{\rm{ }} - 10.683{\rm{ }} - {\rm{ }}0.039{\rm{\, }}^{*}{\rm{ age }}\left( {{\rm{years}}} \right){\rm{ }} + {\rm{ }}0.186{\rm{\, }}^{*}{\rm{ height }}\left( {{\rm{cm}}} \right){\rm{ }} $$$+ {\rm{ }}0.383{\rm{\, }}^{*}{\rm{ weight }}\left( {{\rm{kg}}} \right){\rm{ }} - {\rm{ }}0.043{\rm{\, }}^{*}{\rm{ waist circumference }}\left( {{\rm{cm}}} \right){\rm{ }} $$$- {\rm{ }}0.340$$
Covariates
Based on previous studies and the characteristics of data, we selected covariates a priori, including age, sex, In addition, to assess the independent association of LBM and DNAm with cognition we included BMI as covariate (Lee et al., Reference Lee, Keum, Hu, Orav, Rimm, Willett and Giovannucci2018). We performed within-pair correlation analyses for monozygotic twins; if variables were uncorrelated, they were included as covariates in the model for adjustment (W. Li et al., Reference Li, Zhang, Wang, Wu, Mohammadnejad, Lund, Baumbach, Christiansen and Tan2019).
Reduced Representation Bisulfite Sequencing (RRBS) Analysis
RRBS was used to conduct a whole-genome DNAm analysis on twin blood samples, with details provided in previous studies (J. Liu et al., Reference Liu, Wang, Luo, Duan, Xu, Tian, Chen, Ge and Zhang2024). This analysis identified a total of 551,447 original CpGs. Genomic DNA was digested with MspI to generate CpG-enriched short fragments, followed by selection, bisulfite conversion, and sequencing. Data preprocessing included mapping with Bismark (Krueger & Andrews, Reference Krueger and Andrews2011) and smoothing with BiSeq (Hebestreit et al., Reference Hebestreit, Dugas and Klein2013), setting coverage at the 90th percentile, and excluding CpGs with β values < 0.05 or more than 10 missing observations. After quality control, a total of 248,840 CpGs were available for subsequent analyses. For ease of analysis, β values were transformed into M values using log2. The ReFACTor method was used to adjust for cell type differences, including its top five components as covariates (Jaffe & Irizarry, Reference Jaffe and Irizarry2014; Rahmani et al., Reference Rahmani, Zaitlen, Baran, Eng, Hu, Galanter, Oh, Burchard, Eskin, Zou and Halperin2016).
Statistical Analysis
For continuous variables we described mean (standard deviation, SD) or median (interquartile range, IQR), and for categorical variables, we described number (percentage).
Mediation Analyses
Analyses A: Mediating role of DNAm in the associations between LBM and cognition
A detailed description of the mediation analysis can be found in previous studies (J. Liu et al., Reference Liu, Wang, Luo, Duan, Xu, Tian, Chen, Ge and Zhang2024). The causal inference testing (CIT) method was applied to analyses causal associations and screen for potential mediating variables by fitting four models (Millstein et al., Reference Millstein, Chen and Breton2016). Briefly, the analyses steps were as follows:
-
(1) Model 1: The Generalized Estimating Equation (GEE) was constructed with the ‘geepack’ package in R to examine the relationship between LBM and cognition.
-
(2) Model 2: A genomewide DNA methylation association analysis was conducted to assess the association between LBM and DNAm while controlling for cognition. A Manhattan plot was generated for the results, with multiple testing corrected by calculating the false discovery rate (FDR), defining FDR < 0.05 as genomewide significance. The CpGs with FDR ≥ 0.05 and p < 1×10-5 were considered suggestive, and CpGs with p < 1×10-4 were identified as key sites for further analysis (Juvinao-Quintero et al., Reference Juvinao-Quintero, Sharp, Sanderson, Relton and Elliott2023).
-
(3) Model 3: Under the condition of controlling LBM, the GEE analysis was used to examine the relationship between DNAm of CpGs and cognition. DNAm of CpGs with p < .05 were selected as candidate mediating sites. In the above three CIT models, in order to address the paired structure of twin data, we set the parameter id = family_ID to account for within-twin-pair correlations, while setting the parameter corstr = ‘exchangeable’ to account for between-twin-pair correlations.
-
(4) Model 4: Mediation analysis was conducted with the ‘PROCESS v4.0’ in SPSS 27.0, using LBM as the independent variable and cognition as the dependent variable, with DNAm of CpGs from Model 3 as the potential mediator. After 5000 bootstrap simulations, we calculated average causal mediation effect (ACME), average direct effect (ADE), and total effect (TE). When the ADE and ACME were in opposite directions, we calculated the absolute value of the ACME/ADE ratio (|ACME/ADE|) to report the mediation effect; when they were in the same direction, we calculated the absolute value of the ACME/TE ratio (|ACME/TE|; Wen & Ye, Reference Wen and Ye2014). If the 95% CI of one CpG’s ACME excluded zero, its DNAm was considered a significant mediator in the LBM-cognition association. Based on the estimates of DNAm, single-gene methylation risk scores (MRSs) were calculated for CpGs located in or near (within 40,000 base pairs upstream/downstream of each CpG site) the same gene with mediation effects, as well as a total MRS for all mediating CpGs (Hüls & Czamara, Reference Hüls and Czamara2020). The MRSs were also used as mediating variables for mediation analyses, respectively. The calculation formula for MRS was as follows:
$$MR{S_i} = {w_1}{m_{i1}} + \cdots + {w_k}{m_{ik}}$$
$k$
, the number of preselected CpG sites;
${w_{1,}} \cdots {w_k}$
,the β-values of the association;
${m_{i1}} \cdots {m_{ik}}$
, the methylation level.
Analyses B: Mediating role of LBM in the associations between DNAm and cognition
The analyses steps were nearly consistent with those described above: (1) Model 1: Evaluating the association between cognition and DNAm. (2) Model 2: Evaluating the association between DNAm at each CpG identified in Model 1 and LBM, controlling for cognition. (3) Model 3: Evaluating the association between LBM and cognition, controlling CpGs identified in Model 2. (4) Model 4: Using the ‘PROCESS v4.0’ in SPSS 27.0, with 5000 bootstrap simulations, calculating ACME, ADE, TE, |ACME/ADE | or |ACME/TE|.
Genomic Region Enrichment Analysis
In this study, the CpGs with p < .05 from the association analyses between DNAm and phenotypes (Model 2 and Model 3 in Analyses A, Model 1 and Model 2 in Analyses B) were submitted to the Genomic Regions Enrichment of Annotations Tool (GREAT) for genomic region enrichment annotation to identify gene enrichment regions. In the gene enrichment analysis, FDR < 0.05 was considered significant.
Differentially Methylated Region Analysis
The comb-p method was used to identify differential methylation regions (DMRs) associated with CpGs (P < .05) (Pedersen et al., Reference Pedersen, Schwartz, Yang and Kechris2012; Wang, Li, Jiang et al., Reference Wang, Li, Jiang, Lin, Wu, Wen, Xu, Tian, Li, Tan and Zhang2021; Wang, Li, Wu et al., Reference Wang, Li, Wu, Tian, Duan, Li, Tan and Zhang2021). The DMRs were deemed statistically significant if the Stouffer-Liptak-Kechris (slk) corrected p < .05.
Results
Characteristics of Participants
Table 1 shows the characteristics of participants. A total of 59 twin pairs were included in this study, with 31 male pairs. The median age (IQR) of the participants was 52 (47.0–58.0) years, the mean BMI (SD) was 25.0 (3.7) kg/m2, the median LBM (IQR) was 41.7 (37.5–45.7) kg, and the median cognition score (IQR) was 20.0 (16.0–24.3) points. Serum uric acid (SUA) showed no significant within-pair correlation. Therefore, SUA was included as a covariate in subsequent analyses.
Basic characteristics of participants
Note: The normally distributed quantitative data used Pearson correlation to calculate the twin’s internal correlation coefficient; the non-normally distributed quantitative data used Spearman rank correlation to calculate correlation coefficient.
IQR, interquartile range; BMI, body mass index; LBM, lean body mass; SBP, systolic blood pressure; DBP, diastolic blood pressure; GLU, fasting glucose; CHOL, total cholesterol; TG, triglyceride; SUA, serum uric acid
Analyses A: Mediating Role of DNAm Between LBM and Cognition
Model 1: Association between LBM and cognition
We found that LBM was positively associated with cognition (β = 0.422, p < .001).
Model 2: Association between LBM and DNAm
The associations between LBM and DNAm of 48 CpGs reached genomewide significance (FDR < 0.05), and DNAm of 213 CpGs reached p < 1×10-4 (Table S2, Figure S1). These CpGs were in or near 82 genes, such as PAXIP1-AS2, AGRN, CBX3P4, VSTM2B, H3Y2, SLC26A10, SIM1, GABRB3, GABRA5, and RP11. The strongest association was detected in the CpG within CBX3P4 gene (β = 0.06, FDR = 2.18×10-4).
We identified 22 DMRs significantly associated with LBM (FDR < 0.05) (Table S3). Among these, LBM was negatively associated with methylation levels in 4 DMRs (Figure 1-B, C, D, N) and positively associated with 10 DMRs (Figure 1-E, F, G, H, J, O, P, R, U, V). For the other DMRs, the associations with methylation levels remained unclear (Figure 1-A, I, K, L, M, Q, S, T). Notably, the DMRs located in or near genes such as VSTM2B, GATA5, GABRB3, GABRA5, CNOT6, NPIPB11, TTC40, AGRN, and MRGPRF cover the majority of CpGs listed in Table S2.
Differential methylation patterns of the identified differentially methylated regions for LBM. The dots represent the CpGs. The x-axis shows the position of CpGs on chromosome and the y-axis shows regression coefficients.
Note: BP, base pair; DMR, differentially methylated regions.
The pathways associated with LBM primarily involved aplatelet-derived growth factor binding, ATP synthesis, genes involved in glucagon signaling in metabolic regulation, Interferon-gamma signaling pathway, genes involved in acetylcholine binding and downstream events (Table S4).
Model 3: Association between DNAm and cognition
After adjusting for LBM, we identified significant associations between the DNAm level of 36 CpGs and cognition (p < .05) (Table S5). These CpGs were located in or near the PDGFRB, RP11, COL23A1, HMCN2, MAPK10, CCND3, PAXIP1-AS2, TNFRSF4, CBX3P4, SLC38A10, HCRTR2, ELFN1, ATAD3C, VSTM2B, C14orf39, PLEKHA7, RNU6-955P, and PAX2 genes. The biological pathways associated with cognition include protein folding, nitrogen cycle metabolic process, and others (Table S6).
Model 4: Mediation analysis
We identified six CpGs located in or near the PDGFRB (ADE: 0.5312, ACME: -0.0972), RP11 (ADE: 0.4803−0.4872, ACME: −0.533−0.0463), and PAX2 (ADE: 0.3510, ACME: 0.0831|) genes that could mediate the association between LBM and cognition (Table 2). Additionally, the MRS for the RP11 gene also demonstrated a mediating effect between LBM and cognition (ADE: 0.4857, ACME: −0.0517). However, no significant mediation effect was observed for the total MRS. Except for CpG in PAX2 gene, the ADE values were positive, indicating that LBM has a positive effect on cognition; the ACME values were negative, meaning that DNAm of these CpGs negatively mediate the impact of LBM on cognition.
The results of DNA methylation mediate the association of lean body mass (LBM) and cognition
Note: ADE, average direct effect; ACME, average causal mediation effect; |ACME/ADE|, absolute value of the ratio of mediated effect to direct effect.
Analyses B: Mediating Role of LBM Between DNAm and Cognition
Model 1: Association between DNAm and cognition
We found that the DNAm level of 85 CpGs were significantly associated with cognition, reaching the genomewide level (FDR < 0.05), and DNAm level of 265 CpGs were associated with cognition, reaching p < 1×10−4 (Table S7). These CpGs were in or near 113 genes, including FOXI2, C5orf17, ZNF696, PIK3C2G, SLC22A14, AGPAT2, CPZ, DLGAP4, GPR123, PEX10, and others. The strongest association was observed at a CpG in the PRR12 gene (β = −0.06, FDR = 1.02 × 10−3). Biological pathways related to cognition included platelet-derived growth factor binding, ATP synthesis, negative regulation of interleukin-4 biosynthetic process, circadian clock system, genes involved in glucagon signaling in metabolic regulation (Table S8).
Model 2: Association between DNAm and LBM
For control cognition, there were significant associations between DNAm level of 24 CpGs and LBM (p < .05). These CpGs were in or near DLGAP4, FOXI2, RTN4RL1, TMEM51, PIK3C2G, TRA2B, HSPB7, PS8L2, INC02109, UC5B/MUC5AC, KIAA0232, UBXN6, GPR123, BLM, UNCX, and CIC genes (Table S9). The biological pathways potentially associated with LBM mainly include structural constituent of muscle, central nervous system myelination, genes involved in muscle contraction, and so forth (Table S10).
Model 3: Association between LBM and cognition
LBM was significantly associated with cognition (β: 0.3071–0.4898). In other words, with the increase of LBM, the cognition of participants increased (Table S11)
Model 4: Mediation analysis
LBM mediated the associations of DNAm level of 4 CpGs located in or near to FOXI2 gene with cognition (ADE: 0.7518–1.9279, ACME: 0.2821–0.4496), and DNAm level of one CpGs located in or near to UBXN6 gene with cognition (ADE: -0.2791, ACME: -0.0732) (Table 3).
The results of LBM mediate the association of DNA methylation and cognition
Note: ADE, average direct effect; ACME, average causal mediation effect; |ACME/ADE|, absolute value of the ratio of mediated effect to direct effect.
Discussion
Based on a sample of 59 monozygotic twin pairs, we performed mediation analyses using CIT method. We found that DNAm and MRSs at CpGs could mediate the association between LBM and cognition. LBM could also mediate the association between DNAm at CpGs and cognition. These sites were located at or near the PDGFRB, PR11, PAX2, UBXN6, FOXI2 genes.
When DNAm serves as the mediator, the direction of the indirect effect is opposite to that of the direct effect, which is referred to as a ‘suppression effect’. In other words, the mediation effect of DNAm at CpGs weakened the positive impact of LBM on cognition. Suppression effect has also been observed in other mediation studies on DNAm variation (Briollais et al., Reference Briollais, Rustand, Allard, Wu, Xu, Rajan, Hivert, Doyon, Bouchard, McGowan, Matthews and Lye2021; J. Liu et al., Reference Liu, Wang, Luo, Duan, Xu, Tian, Chen, Ge and Zhang2024). Considering the multifaceted role of DNAm in regulating gene expression, preserving genomic stability, responding to environmental factors, and impacting disease onset and progression, suppression effect appears reasonable.
DNAm of the CpGs, which mediate the association between LBM and cognition, were mainly located in or near to PDGFRB, PR11, PAX2 genes. The functions of these genes are as follows:
The PDGFRB gene encodes the platelet-derived growth factor receptor beta, which is involved in hematopoiesis and vasculogenesis (Hellström et al., Reference Hellström, Kalén, Lindahl, Abramsson and Betsholtz1999; Levéen et al., Reference Levéen, Pekny, Gebre-Medhin, Swolin, Larsson and Betsholtz1994). Decreased PDGFRB expression is associated with primary familial brain calcification, presenting with motor and cognitive impairments (Lenglez et al., Reference Lenglez, Sablon, Fénelon, Boland, Deleuze, Boutoleau-Bretonnière, Nicolas and Demoulin2022). Experimental studies show that PDGFRB signaling participates in glial development, stimulates the migration of oligodendrocytes and neural stem cells, and performs neuroprotective effects in adult mice (Forsberg-Nilsson et al., Reference Forsberg-Nilsson, Behar, Afrakhte, Barker and McKay1998; Smits et al., Reference Smits, Kato, Westermark, Nistér, Heldin and Funa1991). Additionally, PDGFRB promotes the progression of obesity. PDGFRB enhances glycolysis and inflammation, activates microglia, and reduces energy consumption, contributing to the exacerbation of obesity (Okekawa et al., Reference Okekawa, Wada, Onogi, Takeda, Miyazawa, Sasahara, Tsuneki and Sasaoka2024). PDGFRB in pericytes also participates in the formation of new blood vessels in white adipose tissue, increasing and promoting fat accumulation (Onogi et al., Reference Onogi, Wada, Kamiya, Inata, Matsuzawa, Inaba, Kimura, Inoue, Yamamoto, Ishii, Koya, Tsuneki, Sasahara and Sasaoka2017). We hypothesize that an increase in LBM may elevate the methylation level of the PDGFRB gene, suppressing its expression, which could influence cognition by regulating inflammation and neural system activity.
RP11 is a long noncoding RNA (lncRNA). Although lncRNA lacks protein-coding ability, it plays a crucial role as a regulator in the development of obesity, participating in the differentiation of pre-adipocytes and lipogenesis (Z. Chen, Reference Chen2016; Divoux et al., Reference Divoux, Karastergiou, Xie, Guo, Perera, Fried and Smith2014). RP11 is significantly upregulated in people with obesity and promotes the expression of Wnt5β in pre-adipocytes through miR-587 (T. Zhang et al., Reference Zhang, Liu, Mao, Yang, Zhang, Zhang, Guo, Zhan, Xiang and Liu2020). Wnt5β can inhibit the Wnt/β-catenin pathway, thereby promoting fat accumulation within cells. RP11 is also associated with inflammatory responses (Kanazawa et al., Reference Kanazawa, Tsukada, Kamiyama, Yanagimoto, Nakajima and Maeda2005). Overexpression of RP11 increases the expression of heat shock proteins, which in turn triggers inflammation (Asea et al., Reference Asea, Kraeft, Kurt-Jones, Stevenson, Chen, Finberg, Koo and Calderwood2000; M. Liu et al., Reference Liu, Sun, Zhu, Zhu, Deng, Nie, Mo, Du, Huang, Hu, Liang, Wang, Luo, Yi, Zhang, Zhong, Cao and Chen2021). Inflammation is linked to the onset of neurodegenerative diseases such as Alzheimer’s disease (Irwin & Vitiello, Reference Irwin and Vitiello2019). Therefore, an increase in LBM may raise the methylation level of RP11, promoting its expression, leading to cognitive impairment through the increased expression of inflammation-related RNAs.
The protein encoded by the PAX2 gene is an important transcription factor in the development of the central nervous system, involved in controlling the midbrain-hindbrain boundary (Curto et al., Reference Curto, Gard and Ribes2015). PAX2 plays a decisive role in the process by which neural progenitor cells differentiate into neurons or glial cells (Soukkarieh et al., Reference Soukkarieh, Agius, Soula and Cochard2007). Reduced expression of PAX2 may affect intercellular signaling in the brain, and mutations of this gene can lead to neurodevelopmental disorders (Rossanti et al., Reference Rossanti, Morisada, Nozu, Kamei, Horinouchi, Yamamura, Minamikawa, Fujimura, Nagano, Sakakibara, Ninchoji, Kaito, Ito, Tanaka and Iijima2020). Mice with PAX2 gene deletion may exhibit impaired spatial learning and memory due to abnormal synaptic structure (Lv et al., Reference Lv, Wang, Liu, Tang, Lei, Wang and Wei2022). PAX2 deletion may also result in dysfunctional microglia, the main immune cells of the central nervous system, and microglial dysfunction is closely related to neuroinflammation (Streit et al., Reference Streit, Mrak and Griffin2004). Therefore, an increase in LBM may elevate the methylation level of the PAX2 gene, promoting its expression and thereby protecting cognition by maintaining nervous system function.
LBM mediates the associations between DNAm of the CpGs, which were mainly located in or near to UBXN6 and FOXI2 genes, and cognition. The functions of these genes are as follows:
The UBXN6 gene encodes a protein with a UBX domain, which negatively regulates ATPase activity (Trusch et al., Reference Trusch, Matena, Vuk, Koerver, Knævelsrud, Freemont, Meyer and Bayer2015). UBXN6 is involved in the clearance of damaged lysosomes, with lysosomal dysfunction being associated with neurodegenerative diseases such as amyotrophic lateral sclerosis (ALS) and frontotemporal dementia (FTD) (Papadopoulos et al., Reference Papadopoulos, Kirchner, Bug, Grum, Koerver, Schulze, Poehler, Dressler, Fengler, Arhzaouy, Lux, Ehrmann, Weihl and Meyer2017; Root et al., Reference Root, Merino, Nuckols, Johnson and Kukar2021). Inclusion body myopathy with Paget’s disease, characterized by adult-onset muscle weakness, is also linked to UBXN6. In brain tissue of ALS and FTD patients, UBXN6 has been found to closely associate with poly-GA aggregates, which have neurodegenerative toxicity (F. Liu et al., Reference Liu, Morderer, Wren, Vettleson-Trutza, Wang, Rabichow, Salemi, Phinney, Oskarsson, Dickson and Rossoll2022). UBXN1 and UBXN6 function as adaptor proteins, recognizing misfolded proteins in the brain and degrading via the proteasome or autophagosome (Ganji et al., Reference Ganji, Mukkavalli, Somanji and Raman2018; Papadopoulos et al., Reference Papadopoulos, Kirchner, Bug, Grum, Koerver, Schulze, Poehler, Dressler, Fengler, Arhzaouy, Lux, Ehrmann, Weihl and Meyer2017). Misfolded proteins in the brain are significant pathological feature of dementia (Hartl, Reference Hartl2017). Therefore, increased methylation of the UBXN6 gene may suppress its expression, reducing LBM and then potentially impairing cognition.
FOXI2 is widely expressed in the neural layers of the brain and retina, playing an important role in brain development (Wijchers et al., Reference Wijchers, Hoekman, Burbach and Smidt2005). FOXI2 is a key regulatory factor of AgRP neuron activity, which helps maintain whole-body energy homeostasis (Joly-Amado et al., Reference Joly-Amado, Denis, Castel, Lacombe, Cansell, Rouch, Kassis, Dairou, Cani, Ventura-Clapier, Prola, Flamment, Foufelle, Magnan and Luquet2012; Thomas & Xue, Reference Thomas and Xue2018). Overexpression of FOXI2 increases food intake and reduces energy metabolism by inducing AgRP gene expression, leading to impaired glucose tolerance, reduced insulin sensitivity, and obesity (Fan et al., Reference Fan, Sheng, Guo, Qiao, Jin, Tan, Gao, Zhang, Dong, Zhang, Li, Shen, Liao and Chang2022). Therefore, we hypothesize that increased methylation of the FOXI2 gene may reduce its expression, subsequently suppressing AgRP gene expression, leading to an increase in LBM and then protecting cognition.
There are several strengths in this study. First, we utilized a monozygotic twin design with trait discordance, providing robustness in examining epigenetic variation associated with complex diseases. Second, this study is the first to explore the mediating role of DNAm in the association between LBM and cognition, as well as the mediating role of LBM in the association between DNAm and cognition. Third, this study utilized the GREAT tool for enrichment analysis, encompassing not only traditional pathways but also functional annotations, including GO Molecular Function, GO Biological Process, and other ontology categories. Lastly, given the differences in genetic composition, environmental factors, and lifestyles among various ethnic groups of worldwide, this study contributes to illustrating the potential mechanisms underlying cognitive decline in the Han Chinese population.
This study also has several limitations. First, although multiple confounding factors have been adjusted in the mediation analysis, unknown confounders may influence the results. Second, the sample size of this study was relatively small. But based on the result of simulations (W. Li et al., Reference Li, Christiansen, Hjelmborg, Baumbach and Tan2018), the sample size of 59 monozygotic twins’ pairs can achieve 80% statistical power. Third, although multiple variables have been adjusted in this study, the effects from other potential confounding factors cannot be fully controlled. Fourth, there may be collider bias, but the CIT method used in this study, and the associations between LBM, DNAm, and cognition found in this study were consistent with previous studies, which can effectively control collision bias.
Conclusions
This study found that the DNAm of CpGs, which are in or near to PDGFRB, PR11, and PAX2 genes, can partially mediate the association between LBM and cognition. Additionally, LBM can also partially mediate the associations of the DNAm of CpGs in or near to UBXN6 and FOXI2 genes with cognition. The results revealed complex interactions among DNAm, LBM, and cognition, offering new insights into the potential molecular mechanisms underlying the pathogenesis of cognitive impairment.
Supplementary material
To view supplementary material for this article, please visit https://doi.org/10.1017/thg.2025.10014.
Author contributions
Conceptualization: DFZ and WJW. Data curation: DFZ, TW and HHL. Formal analysis: HHL and TW. Investigation: TW, HHL and XCT. Methodology: HHL and TW. Project administration: DFZ and WJW. Resources: DFZ, WJW. Supervision: DFZ, WJW. Visualization: HHL, TW and XCT. Writing — original draft: HHL. Writing — review and editing: DFZ, WJW. The authors read and approved the final manuscript.
Funding
This work was supported by the National Natural Science Foundation of China (82073641).
Competing interests
The authors have no relevant financial or nonfinancial interests to disclose.
Data availability
The datasets generated during and/or analyzed during the current study are available from the corresponding author on reasonable request.
Ethics approval
This study conformed to the declaration of Helsinki, and all participants provided written informed consent. This study was approved by the Regional Ethics Committee of the Qingdao Center for Disease Control (Approval No.: 2012–01; Date: 2012-01-20).
Consent to participate
Informed consent was obtained from all participants included in the study.
