2. Genes Associated with Differentially Methylated CpGs

Individual CpG sites that were differentially methylated were annotated to many of the genes identified as differential in the analysis of gene methylation. cg15613340 in *protocadherin beta 5* (*PCDHB5*) was significantly hypomethylated in vegans in the analysis of methylation of all sites and of the first exon, with a fold change of 0.81 (analyses with SmartSVA), and cg25026992 in *protocadherin beta 7* (*PCDHB7*) was similarly hypomethylated (fold change, 0.81) in analysis of the 3 UTR region (analyses without SmartSVA), representing the most marked hypomethylation observed among CpG sites. The most marked hypermethylation was that of Cg07967210 in *SNF8 subunit of ESCRT-II* (*SNF8*), identified in the analysis of promoter methylation (fold change, 1.17) (Supplementary Tables S5–S8).



Some CpG sites represented in more than one gene region. 3 Number of CpG sites estimated to show non-null differences in methylation. 4 Fold change represents ratio of the mean methylation of vegans to that of non-vegetarians for differentially methylated (hypomethylated or hypermethylated) sites for a given region. This is averaged across significant sites.

1 2

#### **4. Discussion**

DNA methylation is one mechanism that links lifestyle with genomic alterations. The goal of this study was to determine if habitual vegan and non-vegetarian dietary patterns differentially influenced DNA methylation. Using a permutation-based adapted Storey et al. [32] approach, and considering various genomic regions, we could detect a modest number of genes that differed significantly in their methylation status between vegans and non-vegetarians, noting that much larger proportions of apparently differentially methylated CpG sites and genes are in fact present, and should be detected with larger samples.

The majority of differentially methylated genes detected were hypomethylated in vegans, both when considering gene methylation overall, as well as in various gene regions. The greatest differences in gene methylation were found when considering the gene body. The region-specific differences in the number of detected differentially methylated genes are in part a reflection of the nonuniformity in the distribution of probes selected for the BeadChip array (i.e., a large number of CpG sites were present in the gene body). However, there were fewer observed and predicted non-null differentially methylated genes defined by the TSS1500 and CpG island regions, in spite of greater total numbers of genes with CpG sites in these regions, relative to the gene body. The gene body may therefore be the most susceptible to diet-induced alterations in methylation. Overall, we have estimated larger proportions of differentially methylated, non-null genes (6% of all genes) and CpG sites (9% of all CpG sites) to be detected with sufficiently large samples, using our adapted Storey approach. It should be noted that results from our permutation-based adapted Storey approach showed strong agreement with results from the simpler Benjamini–Hochberg approach for false discovery.

There is little other published evidence on the influence of plant-based dietary patterns on DNA methylation. Global methylation (LINE1) has been associated with consumption of fruits and vegetables or flavonoids in dietary intervention studies [41,42], as well as with energy restriction in overweight participants [43,44] although the latter has not been consistently observed [45]. It is unclear how a vegan diet impacted global (overall) methylation in the current study, given our array-based approach. However, of the significantly hypermethylated CpG sites, the majority were located in the intergenic region, which houses many transposable elements, including LINE1. Our findings show some consistency with those reported by Perfilyev et al., where high fat feeding (both saturated and polyunsaturated combined) was associated with hypermethylation of genes (>99%) [27]. These findings seem to complement our results as vegans who have lower fat consumption [9] and lower proportions of saturated fatty acids in adipose tissue [10], showed decreased gene methylation overall. However, the differential patterns of methylation between vegans and non-vegetarians in our study was not merely due to differences in fat, as our findings were only slightly attenuated when adjusted for total/saturated fat. Additionally, associations persisted when controlling for BMI, which could also be thought of as a surrogate of long-term energy intake and energy balance.

Many genes differentially methylated in the analysis of overall methylation of all genes have roles in RNA transport or ribosome assembly/regulation of translation (*RPL38*, *TUFM* [46,47]), as well as lysosome regulation (HPS4 [48]), protein degradation or ubiquitination (*DCAF15*, *PSMF1* [49,50]), cytokine or B cell receptor signaling (*CMTM7*, *IL36A* [51,52]), and metabolism (*STRA6*, *COX7A2L*, *GALNT12* [53–56]), among other pathways. Alterations in many of these processes impact protein synthesis and may have pathophysiological implications [57,58].

We have highlighted methylation differences in the promoter region particularly, as such alterations may have a greater impact on gene transcription. *METTL1* consistently showed hypomethylation in all relevant analyses—i.e., analyses of methylation of genes defined by CpGs in associated promoter regions, as well as of overall methylation of all genes, this being in analyses including or excluding SmartSVA variables, and after adjustment for false discovery. *METTL1* encodes a methyltransferase that promotes methylguanine methylation of RNA, including miRNAs such as let7, which results in processing of the miRNA [59]. Let-7 miRNA has roles in regulating cell growth and metabolism, and is also considered a tumor suppressor, showing downregulation in a number of cancers [60]. *METTL1* along with *NEIL2,* encoding an enzyme involved in DNA repair [61], were significantly hypomethylated in CpG islands within the promoter independent of SmartSVA variables. Hypermethylation of CpG islands within the promoter is associated with silencing of tumor suppressor genes [62–65]. Thus, hypomethylation and consequent increases in these genes might be associated with sustained expression of tumor suppressors, and thus cancer prevention.

Besides METTL1 and NEIL2, several of the differentially methylated genes identified by analysis of associated promoter methylation, or of all genes, have roles in cancer biology. For example, *LIN28B* regulates let-7 miRNA [66], *CMTM7* is potential tumor suppressor, possibly silenced by promoter methylation [49,67]), and *ARGHDIB*, a regulator of Rho family signaling, can be upregulated in tumors [68]. Interestingly, *GALNT12*, *DDT*, and *ARNT2* have been found to be over-represented in cancer, including colorectal cancer among others [69–72], with ARNT2 showing promoter hypermethylation in hepatoma cells [73]. We have found that vegans and other vegetarians in AHS-2 have lower risk of certain types of cancer and other chronic diseases [2,3,6]. Thus, DNA methylation alterations in genes with relevance to cancer could help explain some of the differences in cancer outcomes between dietary groups in the AHS-2 cohort. However, it is unclear if transcriptional changes are associated with any of the observed methylation alterations.

The observed trend towards hypomethylation of genes in vegans could be explained in part by inhibition of activity of DNMT enzymes by polyphenols and various secondary plant metabolites [14,74]. We have shown that vegetarians in the AHS-2 cohort have significantly increased consumption of plant-based foods including fruits, vegetables, legumes, nuts, and seeds [9], which are high in polyphenols. Furthermore, we recently demonstrated that vegetarians, particularly vegans, have significantly higher levels of bioactive compounds and phytochemicals-enterolactone, isoflavones, carotenoids, as well as total omega-3 fatty acids (attributable to alpha linolenic acid) in plasma, urine, and adipose tissue, many of which could theoretically modulate methylation [75]. Additionally, methyl donor compounds, folate, choline, methionine and cofactors (vitamin B), which are obtained from various foods, may influence both gene-specific and global DNA methylation, as demonstrated through observational and intervention studies; however, activity may be context, dose, or tissue dependent [13,30,76,77]. It is not clear if any of these methyl donor compounds differ between vegans and non-vegetarians in the current study. Thus, gene-specific and genome-wide alterations in methylation may be determined by a complex interplay of methyl donors, micronutrients, polyphenols, macronutrient composition, and inflammatory status, among other possible factors, including physical activity and stress.

Methylation activity of CpGs within the same gene, and particularly the same region, may be coordinated. This correlation of CpG sites could partly explain why our analysis of gene methylation revealed greater differential alterations than analysis of individual CpG sites, although a much greater number of sites are estimated to be non-null and detectable with a larger sample. Findings from our study suggest that methylation alterations associated with dietary patterns (and likely other environmental exposures) may be characterized by coordinated methylation of CpG sites in a specific gene and within select regions of the gene. This approach of analyzing CpGs within select regions after annotation to their respective genes is, in general, similar to the region-centric approach taken by Bacalini et al. involving the grouping together of probes mapping to the same island or gene [78].

The effect size of methylation differences was low overall, with average regional methylation differences in the range of 2–4% for significant genes and 4–9% for significant CpG sites (at FDR < 0.05), though fold changes were much larger for select genes or CpG sites. Effect sizes of low magnitude (≤1–5%) have been previously reported in dietary studies. Besides the high fat feeding study by Perfilyev et al. [27], interventions examining alterations associated with a Mediterranean diet or diet enriched in flavonoids and isothiocyanates also found very subtle but significant changes (~1% methylation difference in LINE1) [41,44]. Smaller changes in methylation tend to be disregarded [79,80], but the physiological relevance of such small effects is unclear. It is possible that such changes over the long-term contribute to sustained, physiologically meaningful epigenetic differences. These "small" changes may translate into permanent changes in gene expression, and consequently phenotype, and be inherited by daughter cells [81]. The location of methylation alterations also holds much relevance, as methylation of one site can contribute to gene silencing if the methylated site blocks binding of a transcription factor [82,83].

This is the first study to our knowledge analyzing differences in DNA methylation between vegans and non-vegetarians. The AHS-2 cohort is unique in its relatively large number of vegans (~9% of cohort), thereby enabling such a study, although it should be noted that AHS-2 non-vegetarians tend to have lower meat consumption relative to the general population. The examination of habitual, long-term dietary patterns is a major strength, as epigenetic marks are more stable, reflecting long-term dietary conditions. We have shown that the majority of cohort members remain in the same diet group from one decade to the next [84]. Because of the habitual, a priori-classified dietary patterns, there was likely minimal measurement error in the classification of diet group. Additionally, our study was strengthened by leveraging multiple analytical approaches, comparing the SmartSVA method which adjusts for surrogate variables and thus unwanted variation, with linear regression models not including these surrogate variables but adjusting for other known technical variations.

Our study has some noteworthy limitations. Methylation alterations are best paralleled with gene expression data. In the absence of such data there are limitations in the interpretation of our findings. As mentioned, it is not clear how other dietary or lifestyle influences (exercise, methyl donors) may have altered results. Furthermore, we do not know how methylation alterations in leukocytes correlate with those in other tissues, although there is evidence that methylation patterns or alterations in blood may be reflective of patterns in other tissues [85–87]. Additionally, the two dietary groups examined were somewhat heterogeneous in terms of consumption of animal- and plant-based foods. For example, it is not clear how methylation patterns differ comparing individuals with very high consumption of red or processed meat with individuals following a diet comprised largely of fruits, vegetables, and whole plant foods.

#### **5. Conclusions**

In summary, modest specific differences in methylation of genes and CpG sites were detected, comparing vegans and non-vegetarians, with clear indication that many more such differences exist, but are yet to be specifically identified with larger studies. This study thus lays the foundation for the identification of transcriptional alterations and molecular functions associated with these diet-influenced methylation patterns.

**Supplementary Materials:** The following are available online at http://www.mdpi.com/2072-6643/12/12/3697/s1. Figure S1: Overlap of promoter with gene- or island-related regions, Table S1: Estimated non-null and observed differentially methylated genes (FDR < 0.05) summarized according to genic/intergenic region or in relation to CpG islands in vegans relative to non-vegetarians (without SmartSVA method), Table S2: Estimated non-null and observed differentially methylated CpG sites (at FDR < 0.05) summarized according to genic/intergenic region or in relation to CpG islands in vegans relative to non-vegetarians (without SmartSVA), Table S3: Genes differentially methylated at FDR < 0.05 according to genic/intergenic region, Table S4: Genes differentially methylated at FDR < 0.05 in island-related and promoter regions, Table S5: CpGs differentially methylated at FDR < 0.05 according to genic/intergenic region (based on SmartSVA method), Table S6: CpGs differentially methylated at FDR < 0.05 in island-related and promoter regions (based on SmartSVA method), Table S7: CpGs differentially methylated at FDR < 0.05 according to genic/intergenic region, Table S8: CpGs differentially methylated at FDR < 0.05 in island-related and promoter regions.

**Author Contributions:** The authors' responsibilities were as follows: Conceptualization and design, G.E.F., F.L.M., K.S. and P.D.-H.; Methodology, A.M., V.F., G.E.F., F.L.M., C.W., K.S. and X.C.; Data analysis, A.M., G.E.F. and F.L.M.; Writing—Original Draft Preparation, F.L.M. and G.E.F.; Data curation, A.M., G.E.F. and V.F.; Writing—Review and Editing, G.E.F., F.L.M., K.S., M.J.O., P.D.-H., A.M. and C.W. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research was funded by Ardmore Institute of Health, the National Institute of Health (National Cancer Institute, R01 CA094594, National Institute of Aging, 1R01AG026348, and by Loma Linda University through Grants for Research and Partnerships awards (2120211, 2130279, 2170322).

**Conflicts of Interest:** The authors declare no conflict of interest. The funders had no role in the design of the study; in the collection, analyses, or interpretation of data; in the writing of the manuscript, or in the decision to publish the results.
